Papers with LoRA
PAI-Diffusion: Constructing and Serving a Family of Open Chinese Diffusion Models for Text-to-image Synthesis on the Cloud (2024.acl-demos)
Copied to clipboard
| Challenge: | Existing diffusion models fail to address the challenges of generating high-quality images from textual descriptions due to its large vocabulary size and complex character relationships. |
| Approach: | They propose a framework that integrates Chinese diffusion models with Alibaba Cloud's Platform for AI and enables the generation of contextually relevant images. |
| Outcome: | The proposed framework integrates with Alibaba Cloud’s Platform for AI, providing accessible and scalable solutions. |
A Persona-Aware LLM-Enhanced Framework for Multi-Session Personalized Dialogue Generation (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing personalized dialogue models focus on dialogue history and personality information, reducing the responses’ consistency. |
| Approach: | They propose a Persona-Aware LLM-enAnCEd(PALACE) framework that generates responses consistent with dialogue history and personality information across multiple sessions to engage users’ interest in the dialogue. |
| Outcome: | The proposed framework outperforms the state-of-the-art methods in automatic and human evaluation metrics on the MSC and DuLeMon datasets. |
LAD: LoRA-Adapted Diffusion (2025.emnlp-demos)
Copied to clipboard
| Challenge: | Autoregressive models dominate text generation but suffer from left-to-right decoding constraints that limit efficiency and bidirectional reasoning. |
| Approach: | They propose a framework for non-autoregressive generation that adapts LLaMA models for iterative, bidirectional sequence refinement using LoRA adapters. |
| Outcome: | The proposed framework adapts LLaMA models for iterative, bidirectional sequence refinement using LoRA adapters. |
Annotation Entropy Predicts Per-Example Learning Dynamics in LoRA Fine-Tuning (2026.acl-srw)
Copied to clipboard
| Challenge: | Annotator disagreement on tasks like natural languageinference (NLI) reflects genuine linguistic ambiguity, but most fine-tuning recipes treat every example as equallylearnable. |
| Approach: | They ask whether annotator disagreement on tasks like natural languageinference (NLI) reflects genuine linguistic ambiguity. |
| Outcome: | The proposed method predicts learning dynamics on contested examples under LoRA. |
Semantic Contrastive Adaptation for Multimodal Figurative Language Understanding (2026.acl-srw)
Copied to clipboard
| Challenge: | Existing models for understanding figurative language in images perform well on literal recognition but fail on multimodal figurativ benchmarks. |
| Approach: | They propose a model that adapts to idiomatic and figurative language using literal alignment bias rather than limited model capacity. |
| Outcome: | The proposed model generalizes across five idiom-rich languages despite being trained on English supervision. |
Run LoRA Run: Faster and Lighter LoRA Implementations (2025.acl-industry)
Copied to clipboard
| Challenge: | Existing studies on low-rank adapter training use the default chain of operations while calculating the output. |
| Approach: | They propose a framework that allows for efficient LoRA implementations by introducing low-rank adapters to linear layers and selecting the best forward and backward graphs based on FLOPs and time estimations. |
| Outcome: | The proposed framework significantly improves the speed of neural network training and fine-tuning with low-rank adapters. |
PRILoRA: Pruned and Rank-Increasing Low-Rank Adaptation (2024.findings-eacl)
Copied to clipboard
| Challenge: | Several approaches to parameter-efficient fine-tuning have been proposed . low-rank Adaptation (LoRA) does not consider the varying importance of each layer . |
| Approach: | They propose a method that allocates a different rank for each layer and performs pruning throughout the training process. |
| Outcome: | The proposed method is based on eight GLUE benchmarks and is currently the state of the art. |
Towards Robust and Efficient Federated Low-Rank Adaptation with Heterogeneous Clients (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for federated fine-tuning for Large Language Models suffer from performance degradation at low ranks in heterogeneous data settings. |
| Approach: | They propose a low-rank adaptive model with Alternating freeze and Adaptive rank selection which reduces the number of uploaded parameters by 99.8% . |
| Outcome: | The proposed low-rank Adaptation maintains robustness even under extreme heterogeneity and low rank conditions while preserving communication efficiency. |
Probing Out-of-Distribution Robustness of Language Models with Parameter-Efficient Transfer Learning (2023.starsem-1)
Copied to clipboard
| Challenge: | Pre-trained language models (PLMs) are gaining popularity on many benchmarks, but it is uncertain whether they can handle inputs that have been distributionally shifted. |
| Approach: | They evaluated various PETL techniques to detect out-of-distribution changes as the size of the PLM grows or the transfer methods are altered. |
| Outcome: | The proposed methods can detect out-of-distribution changes as the size of the PLM grows or the transfer methods are altered. |
Parameter-efficient Tuning for Large Language Model without Calculating Its Gradients (2023.emnlp-main)
Copied to clipboard
| Challenge: | Recent parameter-efficient tuning methods can only save 30% of training memory . gradient computation and backpropagation are still necessary for these methods . |
| Approach: | They propose a parameter-efficient tuning method that can be used to fine-tune large language models without calculating gradients. |
| Outcome: | The proposed method saves 30% of training memory and improves performance on large language models. |
On-device System of Compositional Multi-tasking in Large Language Models (2025.emnlp-industry)
Copied to clipboard
Ondrej Bohdal, Konstantinos Theodosiadis, Asterios Mpatziakas, Dimitrios Filippidis, Iro Spyrou, Christos Zonios, Anastasios Drosou, Dimosthenis Ioannidis, Kyenghun Lee, Jijoong Moon, Hyeonmok Ko, Mete Ozay, Umberto Michieli
| Challenge: | Existing approaches to generative AI for large language models struggle when executing complex tasks simultaneously. |
| Approach: | They propose a novel approach tailored specifically for compositional multi-tasking scenarios . they add a learnable projection layer on top of the combined summarization and translation adapters. |
| Outcome: | The proposed approach performs well and is fast in both cloud-based and on-device implementations. |
Towards Federated Low-Rank Adaptation of Language Models with Rank Heterogeneity (2025.naacl-short)
Copied to clipboard
| Challenge: | Low-rank adaptation (LoRA) is an efficient alternative to full-weight adaptation in federated fine-tuning of language models, significantly reducing computational costs. |
| Approach: | They propose a low-rank adaptation method that freezes original weights and trains only the update parametrized as a product of two low-ranked matrices. |
| Outcome: | The proposed method accelerates convergence and enhances the global model’s predictive performance. |
LoRA-MGPO: Mitigating Double Descent in Low-Rank Adaptation via Momentum-Guided Perturbation Optimization (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) adapts large language models by training only a small fraction of parameters, but as the rank of the low-rank matrices increases, LoRA exhibits an unstable “double descent” phenomenon, which delays convergence and impairs generalization by causing instability due to the attraction to sharp local minima. |
| Approach: | They propose a framework that incorporates Momentum-Guided Perturbation Optimization (MGPO) MGPO stabilizes training dynamics by mitigating double descent phenomenon and guiding weight perturbations using momentum vectors from the optimizer’s state. |
| Outcome: | The proposed framework improves performance on natural language understanding benchmarks and shows that it improves convergence and generalization. |
ALoRA: Allocating Low-Rank Adaptation for Fine-tuning Large Language Models (2024.naacl-long)
Copied to clipboard
| Challenge: | Low-rank adaptation (LoRA) has demonstrated commendable performance as a popular method . however, it is implemented with a fixed intrinsic rank that might not be ideal for downstream tasks. |
| Approach: | They propose a method that estimates the importance score of each LoRA rank and prunes abundant LoRA ranks to improve performance. |
| Outcome: | The proposed method outperforms baselines on a variety of tasks with comparable parameters. |
R-LoRA: Randomized Multi-Head LoRA for Efficient Multi-task Learning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) improves performance in multi-task learning by diversifying the head matrices through Multi-Head Dropout and Multi-head Random Initialization. |
| Approach: | They propose a low-rank adaptive approach to fine-tune large language models by approximating weight updates through low-ranked matrices. |
| Outcome: | The proposed approach improves performance in multi-task learning while reducing memory usage and training time. |
Plasticity vs. Rigidity: The Impact of Low-Rank Adapters on Reasoning on a Micro-Budget (2026.eacl-srw)
Copied to clipboard
| Challenge: | Recent advances in mathematical reasoning typically rely on massive scale . yet, can strong reasoning capabilities be induced in small language models under extreme constraints? |
| Approach: | They train small language models with a single GPU for under 24 hours . they find that adapters unlock significant plasticity in standard instruction-tuned models . |
| Outcome: | The proposed model training on a single GPU (48GB) achieves 40% Pass@1 on AIME 24 (an 11.1% improvement over baseline) the model training results show that the adapter capacity and initialization are critical factors. |
FanLoRA: Fantastic LoRAs and Where to Find Them in Large Language Model Fine-tuning (2024.emnlp-industry)
Copied to clipboard
| Challenge: | Lowrank adaptation and its variants introduce significant latency in multi-tenant settings, hindering their applications in the industry. |
| Approach: | They propose a framework to fine-tune LoRA modules on a large-scale instruction tuning dataset. |
| Outcome: | The proposed framework outperforms existing PEFT methods and significantly reduces inference latency. |
Multimodal Instruction Tuning with Conditional Mixture of LoRA (2024.acl-long)
Copied to clipboard
| Challenge: | Multimodal Large Language Models (MLLMs) have demonstrated proficiency in diverse tasks across different domains. |
| Approach: | They propose a method that integrates multimodal instruction tuning with Conditional Mixture-of-LoRA. |
| Outcome: | Experimental results show that MixLoRA outperforms LoRA with the same or higher ranks . MLLMs have demonstrated remarkable proficiency in diverse tasks across domains . |
Adapters Selector: Cross-domains and Multi-tasks LoRA Modules Integration Usage Method (2025.coling-main)
Copied to clipboard
| Challenge: | Parameter-Efficient fine-tuning (PEFT) adapts large language models to specific domains by updating only a small portion of the parameters. |
| Approach: | They propose a framework for better integrating usage of multiple adapters by training a middleman adapter to select the appropriate adapter for inference. |
| Outcome: | The proposed framework can perform cross-domain multi-tasks effectively through the utilization of a compact model in combination with multiple LoRA modules. |
Preserving Pre-trained Representation Space: On Effectiveness of Prefix-tuning for Large Multi-modal Models (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Large multi-modal models (LMMs) are revolutionizing the way machines interact with the world, unlocking new possibilities across multi-dimensional applications. |
| Approach: | They propose a parameter-efficient fine-tuning strategy that combines both . they find that parameter tuning methods distort the feature representation space . |
| Outcome: | The proposed strategy preserves representation space while limiting performance on downstream tasks. |
PILLOW: Enhancing Efficient Instruction Fine-tuning via Prompt Matching (2023.emnlp-industry)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) has been used to adapt Large Language Models to a variety of tasks, but it requires substantial computational resources to perform. |
| Approach: | They propose a low-rank adaptive learning approach that leverages LoRA's in-context learning capability through prompt matching via reinforcement learning in resource-constrained environments. |
| Outcome: | The proposed model improves LoRA performance on evaluation metrics and utilises consumer-grade GPU resources. |
Synthetic Data Fine-Tuning for Effective Team Formation in Enterprises (2026.eacl-industry)
Copied to clipboard
| Challenge: | Existing algorithms for semantic search fine-tune text embeddings to retrieve and rank documents . word embedders allow search systems to measure semantic similarity between vectors . |
| Approach: | They evaluate the effectiveness of synthetic data fine-tuning for Semantic Search in a real-world Enterprise Team Formation problem scenario. |
| Outcome: | The proposed model outperforms existing models on a human-curated dataset. |
Memory-Efficient Backpropagation for Fine-Tuning LLMs on Resource-Constrained Mobile Devices (2025.emnlp-industry)
Copied to clipboard
| Challenge: | Existing work on memory-efficient on-device fine-tuning of large language models with backpropagation has focused on approximating gradients with zeroth-order optimization (ZO). |
| Approach: | They propose a memory-efficient implementation of backpropagation on mobile devices that allows flexible trade-offs between memory usage and compute time while converging faster. |
| Outcome: | The proposed method can fine-tune LLMs with backpropagation using less than 1GB of memory while achieving better performance than the baseline. |
QDyLoRA: Quantized Dynamic Low-Rank Adaptation for Efficient Large Language Model Tuning (2024.emnlp-industry)
Copied to clipboard
Hossein Rajabzadeh, Mojtaba Valipour, Tianshu Zhu, Marzieh Tahaei, Hyock Kwon, Ali Ghodsi, Boxing Chen, Mehdi Rezagholizadeh
| Challenge: | Existing methods to fine tune large language models require huge memory, limiting the choice to acquire Larger models. |
| Approach: | They propose an efficient quantization approach for dynamic low-rank adaptation that can efficiently fine tune large language models on a set of pre-defined LoRA ranks. |
| Outcome: | The proposed method outperforms QLoRA and is competitive to QLouRA and outperformed when employing its optimal rank. |
GenKnowSub: Improving Modularity and Reusability of LLMs through General Knowledge Subtraction (2025.acl-short)
Copied to clipboard
| Challenge: | Large language models (LLMs) struggle with zero-shot generalization due to entanglement of general knowledge and task-specific adaptations. |
| Approach: | They propose a modular framework that disentangles general knowledge and adaptations by constructing a library of task-specific LoRA modules alongside a general-domain LoRA. |
| Outcome: | The proposed framework disentangles general knowledge and task-specific adaptations . it generates residual modules that focus more exclusively on task-relevant information . |
LoRA Soups: Merging LoRAs for Practical Skill Composition Tasks (2025.coling-industry)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) is a popular technique for parameter-efficient fine-tuning of Large Language Models. |
| Approach: | They propose to combine LoRA modules to achieve skill composition . they propose to use concatenation of LoRAs to optimize weights for different LoRA training . |
| Outcome: | The proposed model outperforms existing models and data- merging techniques on math-word problems and domain-specialized corpora. |
PARA: Parameter-Efficient Fine-tuning with Prompt-Aware Representation Adjustment (2024.emnlp-industry)
Copied to clipboard
| Challenge: | Existing methods for parameter-efficient fine-tuning excel in the context of single-backbone multi-tenant applications. |
| Approach: | They propose to integrate a lightweight vector generator within each Transformer layer to improve prompt-aware representation adjustment. |
| Outcome: | The proposed method surpasses current benchmarks in terms of performance despite having a similar number of adjustable parameters. |
Frame-Semantic Knowledge Injection for Event-Level Inference in LLMs (2026.acl-short)
Copied to clipboard
| Challenge: | Large language models (LLMs) are fluent but often brittle when interpretation depends on external information. |
| Approach: | They propose a framework that injects frame-semantic knowledge into Large Language Models via LoRA. |
| Outcome: | The proposed framework can generalize beyond surface cues in large language models. |
RoseLoRA: Row and Column-wise Sparse Low-rank Adaptation of Pre-trained Language Model for Knowledge Editing and Fine-tuning (2024.emnlp-main)
Copied to clipboard
| Challenge: | Pre-trained language models have strong generalizability, but fine-tuning involves updating all parameters, rendering full fine-uning resource-intensive. |
| Approach: | They propose a parameter-efficient fine-tuning method that updates all pre-trained parameters during inference. |
| Outcome: | The proposed method outperforms baseline methods on five benchmarks across 20 datasets. |
VIT-Pro: Visual Instruction Tuning for Product Images (2025.naacl-industry)
Copied to clipboard
Vishnu Prabhakaran, Purav Aggarwal, Vishruit Kulshreshtha, Arunita Das, Sahini Venkata Sitaram Sruti, Anoop Saladi
| Challenge: | general-purpose vision-language models struggle to understand and converse about real-world e-commerce product images. |
| Approach: | a new approach is proposed to use large-scale image-text pairs to train a generative VLM for e-commerce product images. |
| Outcome: | The proposed model outperforms general-purpose VLMs on multiple vision tasks in the e-commerce domain. |
BSharedRAG: Backbone Shared Retrieval-Augmented Generation for the E-commerce Domain (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing work adopts separate modules for retrieval and generation, which may be suboptimal since the retrieval task and generation task cannot benefit from each other to improve performance. |
| Approach: | They propose a backbone-shared RAG framework that uses a domain-specific corpus to continuously pre-train a model and then trains two plug-and-play Low-Rank Adaptation modules based on the shared backbone to minimize retrieval and generation losses respectively. |
| Outcome: | The proposed framework outperforms baseline models by 5% and 13% in Hit@3 upon two datasets in retrieval evaluation and by 23% in terms of BLEU-3 in generation evaluation. |
Memory-Efficient Structured Backpropagation for On-Device LLM Fine-Tuning (2026.acl-industry)
Copied to clipboard
| Challenge: | Existing approaches for fine-tuning large language models require a trade-off between exact gradients with high memory and low memory with noisy estimates (MeZO). |
| Approach: | They propose a method which derivates gradients from LoRA's low-rank structure and manually deriving backward passes to exploit the low-level structure. |
| Outcome: | The proposed method reduces peak memory from 361MB to 136MB for Qwen2.5-0.5B, enabling fine-tuning scenarios previously infeasible on memory-constrained devices. |
Minimal Ranks, Maximum Confidence: Parameter-efficient Uncertainty Quantification for LoRA (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) enables parameter-efficient fine-tuning of large language models . standard LoRA lacks mechanisms for uncertainty quantification, leading to overconfident and poorly calibrated models. |
| Approach: | They propose a parameter-efficient Bayesian LoRA method that decomposes weight updates into low-rank matrices. |
| Outcome: | The proposed method achieves strong performance with improved calibration and generalization while maintaining computational efficiency. |
Separate the Wheat from the Chaff: A Post-Hoc Approach to Safety Re-Alignment for Fine-Tuned Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Large language models achieve effective safety alignment at the time of release, but fine-tuning often compromises safety mechanisms. |
| Approach: | They propose a method that performs safety realignment for large language models . they identify unsafe delta parameters from the fine-tuned models and recalibrate the retained parameters . |
| Outcome: | The proposed method improves safety performance on safety benchmarks and jailbreak attacks while maintaining their performance on downstream tasks. |
FedEx-LoRA: Exact Aggregation for Federated and Efficient Fine-Tuning of Large Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for low-rank averaging of LoRA adapters result in inexact updates. |
| Approach: | They propose a method which adds a residual error term to the pre-trained frozen weight matrix to achieve exact updates with minimal computational and communication overhead. |
| Outcome: | The proposed method achieves exact updates with minimal computational and communication overhead, preserving LoRA’s efficiency. |
MoRE: A Mixture of Low-Rank Experts for Adaptive Multi-Task Learning (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent advances in Large Language Models (LLMs) have revolutionized various domains, offering unprecedented performance across numerous tasks. |
| Approach: | They propose a new Mixture of Low-Rank Experts (MoRE) for multi-task PEFT to improve performance of LLMs with fewer parameters. |
| Outcome: | The proposed method improves performance over multiple tasks and no additional inference cost. |
Decoupling Generalization and Adaptation in Meta-Learning for Large Language Models (2026.acl-short)
Copied to clipboard
| Challenge: | Adapting large language models to specific downstream tasks requires multi-step fine-tuning with substantial training data, incurring significant computational overhead. |
| Approach: | They propose a framework that separates learning generalizable initializations and adaptation through dedicated parameter spaces. |
| Outcome: | The proposed framework outperforms existing meta-learning and standard multi-task baselines on common-sense reasoning, mathematics, logic, medical and coding benchmarks. |
LLMs on a Budget? Say HOLA (2025.emnlp-industry)
Copied to clipboard
Zohaib Hasan Siddiqui, Jiechao Gao, Ebad Shabbir, Mohammad Anas Azeez, Rafiq Ali, Gautam Siddharth Kashyap, Usman Naseem
| Challenge: | Current solutions such as quantization, pruning, and Retrieval-Augmented Generation (RAG) offer only partial optimizations and often sacrifice accuracy, speed, or generality. |
| Approach: | They propose an end-to-end optimization framework for efficient LLM deployment . it leverages Hierarchical Speculative Decoding (HSD) for faster inference without quality loss. |
| Outcome: | HOLA delivers +17.6% EMA on GSM8K, +10.5% MCA on ARC, and reduced latency and memory on edge devices like Jetson Nano. |
ORAL: Prompting Your Large-Scale LoRAs via Conditional Recurrent Diffusion (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to low-rank Adaptation (LoRA) are limited in scalability and controllability. |
| Approach: | They propose a conditional recurrent diffusion framework that generates LoRA parameters directly . they integrate model architecture and textual task specifications to generate task-specific parameters . |
| Outcome: | The proposed framework scales to billions-of-parameter LLMs and maintains controllability. |
SIBO: A Simple Booster for Parameter-Efficient Fine-Tuning (2024.findings-acl)
Copied to clipboard
| Challenge: | Recent advances in parameter-efficient fine-tuning (PEFT) techniques allow for adjustments to only a minor fraction of the parameters of large language models. |
| Approach: | They propose a SImple BOoster to enhance parameter-efficient fine-tuning techniques by injecting an initial residual into the model. |
| Outcome: | The proposed model improves performance on 22 benchmark datasets and can be extended to a range of state-of-the-art techniques. |
NLoRA: Nyström-Initiated Low-Rank Adaptation for Large Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Parameter-efficient fine-tuning is essential for adapting large language models (LLMs). However, LoRA suffers from slow convergence and some recent LoRA variants, such as PiSSA, rely on Singular Value Decomposition (SVD) for initialization. |
| Approach: | They propose to introduce a small intermediate matrix between the low-rank matrices (A) and (B) and propose NyströmLoRA (NLoRA) which leverages Nyström-based initialization for SLoRA to improve its effectiveness and efficiency. |
| Outcome: | The proposed approach improves on 5 natural language generation tasks and 8 natural language understanding tasks with minimal parameter overhead. |
Low-Rank Adaptation for Multilingual Summarization: An Empirical Study (2024.findings-naacl)
Copied to clipboard
| Challenge: | Pre-trained Large Language Models have significantly advanced NLP, but their ever-increasing size poses significant challenges for conventional fine-tuning. |
| Approach: | They investigate the potential of Low-Rank Adaptation (LoRA) in multilingual summarization, a task that is challenging and relatively unexplored. |
| Outcome: | The proposed method outperforms full fine-tuning and cross-lingual transfer strategies in multilingual summarization tasks. |
RoZO: Geometry-Aware Zeroth-Order Fine-Tuning on Low-Rank Adapters for Black-Box Large Language Models (2026.eacl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) have demonstrated exceptional performance across a wide range of tasks, yet fine-tuning them efficiently under black-box or memory-constrained settings remains challenging. |
| Approach: | They propose a Riemannian zeroth-order optimization framework that constrains updates to the tangent space of the LoRA manifold. |
| Outcome: | The proposed framework achieves more stable convergence, tighter variance bounds, and superior performance compared to existing ZO methods. |
RIFF: Learning to Rephrase Inputs for Few-shot Fine-tuning of Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Pre-trained Language Models (PLMs) can be fine-tuned for downstream text processing tasks. |
| Approach: | They propose to use paraphrases to enrich the input text of a few-shot model with a Maximum-Marginal Likelihood objective to improve performance. |
| Outcome: | The proposed methods improve performance beyond what can be achieved with parameter-efficient fine-tuning alone. |
Ada-RS: Adaptive Rejection Sampling for Selective Thinking (2026.acl-industry)
Copied to clipboard
Yirou Ge, Yixi Li, Alec M. Chiu, Shivani Shekhar, Zijie Pan, Avinash Thangali, Yun-Shiuan Chuang, Chaitanya Kulkarni, Uma Kona, Linsey Pang, Prakhar Mehrotra
| Challenge: | Large language models are increasingly being deployed in cost- and latency-sensitive settings . chain-of-thought improves reasoning, but it can waste tokens on simple requests . |
| Approach: | They introduce an algorithm-agnostic sample filtering framework for learning selective reasoning . they show that Ada-RS reduces average output tokens by 80% and reducing thinking rate by 5% . |
| Outcome: | The proposed framework reduces output tokens by 80% and thinking rate by 95% on a synthetic tool call-oriented e-commerce benchmark. |
RB-LoRA: Rank-Balanced Aggregation for Low-Rank Adaptation with Federated Fine-Tuning (2026.findings-eacl)
Copied to clipboard
| Challenge: | Low-rank adaptation (LoRA) improves fine-tuning of foundation models by updating only compact adapter matrices . varying client device capabilities lead to different adapter ranks, causing rank heterogeneity that undermines aggregation. |
| Approach: | They propose a rank-balanced aggregation framework that decomposes each update into rank-wise components and aligns them using analytically derived weights. |
| Outcome: | Experiments on language and vision models show that RB-LoRA improves under one and three rounds of communication in federated learning environments. |
SimSCOOD: Systematic Analysis of Out-of-Distribution Generalization in Fine-tuned Source Code Models (2024.findings-naacl)
Copied to clipboard
| Challenge: | Large datasets are increasingly available for pre-training source code models, but obtaining representative training data that fully covers the code distribution for specific downstream tasks remains challenging due to the task-specific nature and limited labeling resources. |
| Approach: | They propose a systematic approach that simulates various OOD scenarios along different dimensions of source code data properties and investigates model behavior under different fine-tuning methodologies. |
| Outcome: | The proposed approach simulates various OOD scenarios along different dimensions of source code data properties and exposes multiple failure modes attributed to OOD generalization issues. |
Time-LlaMA: Adapting Large Language Models for Time Series Modeling via Dynamic Low-rank Adaptation (2025.acl-srw)
Copied to clipboard
| Challenge: | Recent studies have demonstrated that large language models possess robust pattern recognition and semantic understanding capabilities over time series data. |
| Approach: | They propose a time series model that converts time series input into token embeddings and aligns time sequence embeddables with text prompts. |
| Outcome: | The proposed framework achieves the state-of-the-art (SOTA) performance and has potentials for wide industrial usages. |
Not All Directions Matter: Towards Structured and Task-Aware Low-Rank Model Adaptation (2026.acl-long)
Copied to clipboard
Xi Xiao, Chenrui Ma, Yunbei Zhang, Chen Liu, Zhuxuanzi Wang, Yanshu Li, Lin Zhao, Guosheng Hu, Tianyang Wang, Hao Xu
| Challenge: | Low-Rank Adaptation (LoRA) is a key parameter-efficient fine-tuning method . however, its effectiveness is hampered by semantic drift and structural incoherence . |
| Approach: | They propose a low-rank Adaptation framework that tackles semantic drift and structural incoherence by pruning task-irrelevant directions. |
| Outcome: | Experiments on large language models, vision models, and vision models show that the proposed framework outperforms LoRA and advanced dynamic rank allocation and sparsity-based methods. |
Fin-ExBERT: User Intent based Text Extraction in Financial Context using Graph-Augmented BERT and trainable Plugin (2025.emnlp-industry)
Copied to clipboard
| Challenge: | Financial dialogue transcripts pose a unique challenge for sentence-level information extraction due to their informal structure, domain-specific vocabulary, and variable intent density. |
| Approach: | They propose a framework for extracting user intent–relevant sentences from financial service calls. |
| Outcome: | The proposed framework shows strong precision and F1 performance on real-world transcripts . financial transcripts are a challenge due to their informal structure and domain-specific vocabulary . |
FedLFC: Towards Efficient Federated Multilingual Modeling with LoRA-based Language Family Clustering (2024.findings-naacl)
Copied to clipboard
| Challenge: | Existing frameworks for multilingual modeling face communication costs and parameter interference conflicts. |
| Approach: | They propose a communication-efficient federated learning framework with low-rank adaptation and language family clustering for Multilingual Modeling (MM) they maintain the weights of the base model, updating the lightweight Low-rank adapt parameters to minimize communication costs. |
| Outcome: | The proposed model outperforms the baseline models in performance and reduces communication overhead. |
L4Q: Parameter Efficient Quantization-Aware Fine-Tuning on Large Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Quantization-aware PEFT methods have been developed to reduce memory and computational costs associated with large language models. |
| Approach: | They propose a method that integrates Quantization-Aware Training (QAT) with LoRA to reduce memory overhead and improve model accuracy. |
| Outcome: | The proposed method significantly reduces QAT’s memory overhead while preserving the advantage of QAT in producing fully quantized LLMs with high accuracy. |
Parameter-Efficient Fine-Tuning via Circular Convolution (2025.findings-acl)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) has gained popularity for fine-tuning large foundation models, but its intrinsic low-rank characteristic may limit its performance. |
| Approach: | They propose a low-rank adaptive method that uses low-ranked matrices to represent weight changes. |
| Outcome: | The proposed method reduces trainable parameters and mitigates heavy memory consumption associated with full delta matrices by sequentially multiplying mathbf A and mathbb B with the activation. |
LoRAMoE: Alleviating World Knowledge Forgetting in Large Language Models via MoE-Style Plugin (2024.acl-long)
Copied to clipboard
Shihan Dou, Enyu Zhou, Yan Liu, Songyang Gao, Wei Shen, Limao Xiong, Yuhao Zhou, Xiao Wang, Zhiheng Xi, Xiaoran Fan, Shiliang Pu, Jiang Zhu, Rui Zheng, Tao Gui, Qi Zhang, Xuanjing Huang
| Challenge: | Experimental results show that, as the instruction data increases, LoRAMoE can significantly improve the ability to process downstream tasks, while maintaining the world knowledge stored in the LLM. |
| Approach: | They propose a framework that introduces several low-rank adapters and integrates them by using a router network to freeze the backbone model and force a portion of LoRAs to focus on leveraging world knowledge to solve downstream tasks. |
| Outcome: | The proposed framework freezes the backbone model and forces a portion of LoRAs to focus on leveraging world knowledge to solve downstream tasks, to alleviate world knowledge forgetting. |
TeRA: Vector-based Random Tensor Network for High-Rank Adaptation of Large Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) methods have significantly reduced the number of trainable parameters needed in fine-tuning large language models. |
| Approach: | They propose a vector-based random Tensor network for high-Rank Adaptation method that achieves high-rank weight updates while retaining parameter efficiency. |
| Outcome: | The proposed method outperforms existing PEFT methods while keeping low-rank parameters. |
DLIR: Spherical Adaptation for Cross-Lingual Knowledge Transfer of Sociological Concepts Alignment (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for identifying nuanced sociological concepts fail to capture domain-specific subtleties or require extensive parallel data. |
| Approach: | a new approach to aligning nuanced sociological concepts is proposed . a dual-branch LoRA approach captures core semantics and counteracts specific language perturbations. |
| Outcome: | a new approach outperforms baselines on cross-lingual sociological concept retrieval across 10 languages. |
MoSLD: An Extremely Parameter-Efficient Mixture-of-Shared LoRAs for Multi-Task Learning (2025.coling-main)
Copied to clipboard
| Challenge: | LoRA is a key technique for fine-tuning large pre-trained models, yet its performance in multi-task learning scenarios often falls short. |
| Approach: | They propose a mixture-of-shared-LoRAs model with a dropout strategy . they propose to share the upper projection matrix among different experts . |
| Outcome: | The proposed model exhibits excellent performance in both single-task and multi-task scenarios with robust out-of-domain generalization capabilities. |
Beyond Silent Letters: Amplifying LLMs in Emotion Recognition with Vocal Nuances (2025.findings-naacl)
Copied to clipboard
| Challenge: | Recent studies have demonstrated that Large Language Models possess a form of emotional intelligence, capable of interpreting emotional stimuli in text. |
| Approach: | They propose a method that translates speech characteristics into natural language descriptions and integrates them into LLMs to perform multimodal emotion analysis via text prompts. |
| Outcome: | The proposed method outperforms baseline models that require structural modifications on two datasets showing significant improvements in emotion recognition accuracy. |
LoRA Meets Dropout under a Unified Framework (2024.findings-acl)
Copied to clipboard
| Challenge: | Parameter-efficientfinetuning (PEFT) has gained popularity as a lightweight approach for model customization. |
| Approach: | They propose a parameter-efficient dropout method that is overfitting-prone and parameter-freezed. |
| Outcome: | The proposed method is superior to existing methods and compares with transformer-specific methods. |
RST-LoRA: A Discourse-Aware Low-Rank Adaptation for Long Document Abstractive Summarization (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing methods to integrate rhetorical structure theory into long document summarization models are unexplored. |
| Approach: | They propose to integrate rhetorical structure theory into a long document summarization model by explicitly incorporating rhetorical uncertainty into the model. |
| Outcome: | The proposed models outperform the vanilla LoRA and full-parameter fine-tuning models and outperformed previous state-of-the-art methods. |
TaxoLLaMA: WordNet-based Model for Solving Multiple Lexical Semantic Tasks (2024.acl-long)
Copied to clipboard
| Challenge: | Recent studies in Natural Language Processing widely utilize Large Language Models (LLMs) for their capability to store extensive knowledge. |
| Approach: | They propose an LLM-based model that captures lexical-semantic knowledge from WordNet and test it on multiple lexicals. |
| Outcome: | The proposed model achieves 11 SOTA results and 4 top-2 results out of 16 taxonomy-related tasks. |
RuCCoD: Towards Automated ICD Coding in Russian (2025.emnlp-main)
Copied to clipboard
Alexandr Nesterov, Andrey Sakhovskiy, Ivan Sviridov, Airat Valiev, Vladimir Makharev, Petr Anokhin, Galina Zubkova, Elena Tutubalina
| Challenge: | a new dataset for clinical coding in Russian is available for download . human coders must navigate a wide array of medical terminology and time pressures . |
| Approach: | They present a new dataset for ICD coding in Russian, a language with limited biomedical resources. |
| Outcome: | The proposed model improves accuracy on an in-house EHR dataset from 2017 to 2021. |
Training Long-Context LLMs Efficiently via Chunk-wise Optimization (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent advances in long-context large language models have demonstrated superior retrieval quality compared to retrievalaugmented generation (RAG) approaches. |
| Approach: | They propose a memory-efficient training paradigm that partitions lengthy inputs into manageable chunks. |
| Outcome: | The proposed model expands maximum sequence length from 1K to 16K tokens on a single RTX 3090 GPU, while SpaCO achieves accelerated training speed. |
Revisiting LoRA through the Lens of Parameter Redundancy: Spectral Encoding Helps (2025.findings-acl)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) has emerged as a prominent technique for fine-tuning large foundation models. |
| Approach: | They propose a low-rank Adaptation technique that harnesses the expressiveness of spectral bases to re-parameterize LoRA from a sparse spectral subspace. |
| Outcome: | The proposed technique achieves greater efficiency with fewer parameters than baselines on various downstream tasks, including commonsense reasoning, math reasoning, and code generation. |
A Single Linear Layer Yields Task-Adapted Low-Rank Matrices (2024.lrec-main)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) is a parameter-efficient fine-tuning method that updates initial weight matrix W0 with a delta matrix W . |
| Approach: | They propose a method that updates initial weight matrix W0 with a delta matrix W consisting of two low-rank matrices A and B. |
| Outcome: | The proposed method maintains a performance on par with LoRA despite the fact that the trainable parameters of CondLoRA are fewer than those of LoRA. |
Progressive LoRA for Multimodal Continual Instruction Tuning (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to MCIT address Catastrophic Forgetting and Knowledge Transfer (KT) but using a fixed number of shared LoRA blocks across tasks can lead to knowledge interference. |
| Approach: | They propose a framework that uses a fixed number of shared LoRA blocks to reduce knowledge interference. |
| Outcome: | The proposed framework outperforms existing approaches on the latest MCIT benchmark. |
MTIVE: Multi-Task Image Verification Engine Using Vision-Language Models for E-commerce (2026.acl-industry)
Copied to clipboard
| Challenge: | Vision-language models struggle with noisy real-world images and multi-task requirements. |
| Approach: | They propose a curriculum learning framework that adapts vision-language models through three stages . MTIVE uses frozen base weights with stacked LoRA adapters for shared domain knowledge . |
| Outcome: | MTIVE outperforms open-source and proprietary baselines in standard and continual learning settings. |
Applicability Condition Extraction for Therapeutic Drug-Disease Relations (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for identifying conditions under which a drug can be effective are limited . et al., j. n. d., al. c., and dr. m. s., 2005, are not able to identify context-specific conditions for therapeutic drug–disease relations. |
| Approach: | They propose to annotate triples of drugs, diseases, and applicability conditions from biomedical literature. |
| Outcome: | The proposed method outperforms baselines across evaluation settings. |
Leveraging fine-tuned Large Language Models with LoRA for Effective Claim, Claimer, and Claim Object Detection (2024.eacl-long)
Copied to clipboard
| Challenge: | Existing work on identifying claims has focused on sentence level, neglecting supplementary attributes such as the claimer and claim object of the claim. |
| Approach: | They propose a novel approach to detect claims using large language models in natural language understanding and text generation. |
| Outcome: | The proposed approach transforms claim, claimer and claim object detection task into QA setting. |
PRoLoRA: Partial Rotation Empowers More Parameter-Efficient LoRA (2024.acl-long)
Copied to clipboard
| Challenge: | Partially Rotation-enhanced Low-Rank Adaptation (PRoLoRA) is an intra-layer sharing mechanism that circumvents the drawbacks of peer parameter-sharing methods. |
| Approach: | They propose a partially rotation-enhanced low-rank adaptation (PRoLoRA) that shares four components to reduce the cost of LoRA and improves model capacity. |
| Outcome: | Empirical results show that PRoLoRA outperforms LoRA on multiple instruction tuning datasets. |
Recover-LoRA: Data-Free Accuracy Recovery of Degraded Language Models via Low-Rank Adaptation (2025.emnlp-industry)
Copied to clipboard
| Challenge: | Inference optimizations such as quantization, pruning, format and datatype conversion, model export, and serialization can lead to functional degradations in language model task performance. |
| Approach: | They propose a lightweight and dataset-agnostic method to recover model accuracies from quantization, pruning, format and datatype conversion, model export, and serialization errors. |
| Outcome: | The proposed method recovers model accuracies by 5-17% on MHA and GQA models. |
MELoRA: Mini-Ensemble Low-Rank Adapters for Parameter-Efficient Fine-Tuning (2024.acl-long)
Copied to clipboard
Pengjie Ren, Chengshun Shi, Shiguang Wu, Mengqi Zhang, Zhaochun Ren, Maarten Rijke, Zhumin Chen, Jiahuan Pei
| Challenge: | Large language models (LLMs) are the default paradigm for natural language processing (NLP) as the models’ scale and the diversity of tasks increase, fine-tuning becomes infeasible. |
| Approach: | They propose to freeze original pretrained weights and train a group of mini LoRAs with only a small number of parameters and reduce their rank by 8 times . |
| Outcome: | The proposed model uses fewer trainable parameters while maintaining a higher rank, thereby offering improved performance potential. |
AROMA: Autonomous Rank-one Matrix Adaptation (2025.emnlp-main)
Copied to clipboard
| Challenge: | Low-rank adaptation (LoRA) and adaptive low-rank adaption (AdaLoRa) are effective for large language models but are expensive as model sizes escalate into hundreds of billions of parameters. |
| Approach: | They propose a framework that automatically builds up rank-one components with very few trainable parameters that gradually diminish to zero. |
| Outcome: | The proposed framework significantly reduces parameters compared to LoRA and AdaLoRA while maintaining subspace independence. |
LoRAN: Improved Low-Rank Adaptation by a Non-Linear Transformation (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Recent methods for fine-tuning large language models have shown great improvements on a wide range of NLP tasks. |
| Approach: | They propose to introduce a non-linear transformation to improve performance of adapters by introducing a low-rank adaptation to fit the accumulated weight updates. |
| Outcome: | The proposed method outperforms a baseline on SAMSum and 20 Newsgroups tasks and even improves the classification task by 1.95 points when a lower rank is applied. |
LoRAPrune: Structured Pruning Meets Low-Rank Parameter-Efficient Fine-Tuning (2024.findings-acl)
Copied to clipboard
| Challenge: | Low-rank adaption (LoRA) is a low-level pruning method that can be expensive and slow to deploy. |
| Approach: | They propose a low-rank adaption pruning framework that provides an accurate structured pruned model in a memory-efficient manner. |
| Outcome: | The proposed pruning framework reduces perplexity and memory usage by 52.6% on LLaMA and T5 models while reducing memory usage. |
OBLIVIATE: Robust and Practical Machine Unlearning for Large Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) trained over corpora risk memorizing sensitive, copyrighted, or toxic content. |
| Approach: | They propose a framework that removes targeted data while preserving model utility. |
| Outcome: | The proposed framework resists membership inference attacks, minimizes impact on retained data, and maintains robustness across diverse scenarios. |
SOS-LoRA: Static Orthogonal-Subspace Low-Rank Adaptation with Fixed Multi-Scale Scaling (2026.acl-long)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) is a parameter-efficient fine-tuning method for large language models. |
| Approach: | They propose a drop-in extension that reparameterizes a rank-rtot update as a sum of K *static* low-rank experts. |
| Outcome: | Experiments on reasoning and knowledge-intensive benchmarks show consistent gains over matched-budget LoRA. |
Interpreting Arithmetic Mechanism in Large Language Models through Comparative Neuron Analysis (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies have found that arithmetic ability is limited to a few attention heads . existing studies do not elaborate on the mechanisms of these heads or how they influence FFN layers. |
| Approach: | They propose a method that identifies an internal logic chain consisting of four stages from input to prediction. |
| Outcome: | The proposed method improves prediction probabilities by amplifying coefficient scores of FFN neurons related to predictions. |
Convolutional LoRA Aggregation for Unseen Tasks Adaptation (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing LoRA selection methods rely on a few task samples, making it difficult to capture full scope of task-relevant information. |
| Approach: | They propose a framework that selects appropriate LoRA modules and aggregates them using a convolutional LoRA aggregator. |
| Outcome: | The proposed framework bridges the knowledge gap between selected modules and target task . it ensures comprehensive coverage of task-relevant LoRA modules . |
Beyond Full Fine-tuning: Harnessing the Power of LoRA for Multi-Task Instruction Tuning (2024.lrec-main)
Copied to clipboard
Chunlei Xin, Yaojie Lu, Hongyu Lin, Shuheng Zhou, Huijia Zhu, Weiqiang Wang, Zhongyi Liu, Xianpei Han, Le Sun
| Challenge: | Low-Rank Adaptation (LoRA) is a parameter-efficient fine-tuning algorithm for large-scale language models. |
| Approach: | They conduct a systematic study of Low-Rank Adaptation (LoRA) on diverse tasks and rich resources with different learning capacities. |
| Outcome: | The proposed algorithm can achieve remarkable performance in high-resource and multi-task scenarios, even comparable to full fine-tuning. |
STAR: Constraint LoRA with Dynamic Active Learning for Data-Efficient Fine-Tuning of Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing studies show that supervised training is still necessary for complex reasoning tasks. |
| Approach: | They propose a method to integrate uncertainty-based active learning and LoRA to effectively integrate the two methods. |
| Outcome: | The proposed approach outperforms baseline models on three reasoning tasks. |
Interpretable Safety Alignment via SAE-Constructed Low-Rank Subspace Adaptation (2026.acl-long)
Copied to clipboard
| Challenge: | Prior work has shown that safety behaviors are governed by low-rank structures . Low-Rank Adaptation (LoRA) consistently underperforms full fine-tuning and reinforcement learning on safety benchmarks . |
| Approach: | They propose a safety alignment system that disentangles safety-relevant directions into monosemantic features and constructs an interpretable safety subspace from SAE directions. |
| Outcome: | Empirically, the proposed model achieves 99.6% safety rates across multiple model families and scales . low-rank Adaptation consistently underperforms full fine-tuning and reinforcement learning on safety benchmarks compared with previous methods . |
Wanda++: Pruning Large Language Models via Regional Gradients (2025.findings-acl)
Copied to clipboard
Yifan Yang, Kai Zhen, Bhavana Ganesh, Aram Galstyan, Goeric Huybrechts, Markus Müller, Jonas M. Kübler, Rupak Vignesh Swaminathan, Athanasios Mouchtaris, Sravan Babu Bodapati, Nathan Susanj, Zheng Zhang, Jack FitzGerald, Abhishek Kumar
| Challenge: | Existing pruning methods suffer from accuracy degradation without full-model sparsity-aware fine-tuning. |
| Approach: | They propose a pruning framework that uses decoder-block-level regional gradients to improve pruning accuracy. |
| Outcome: | The proposed pruning framework outperforms the state-of-the-art pruning frameworks by utilizing decoder-block-level regional gradients. |
SSMLoRA: Enhancing Low-Rank Adaptation with State Space Model (2025.naacl-long)
Copied to clipboard
| Challenge: | Fine-tuning requires substantial computational resources and is prone to overfitting when applied to small datasets. |
| Approach: | They propose a parameter-efficient fine-tuning method that integrates a State Space Model (SSM) to interconnect low-rank matrices. |
| Outcome: | The proposed method achieves comparable performance to LoRA on the general language understanding evaluation (GLUE) benchmark while using only half the parameters. |
Unlocking the Effectiveness of LoRA-FP for Seamless Transfer Implantation of Fingerprints in Downstream Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | lightweight plug-and-play framework that encodes backdoor fingerprints into LoRA adapters . |
| Approach: | proposed framework encodes backdoor fingerprints into LoRA adapters via constrained fine-tuning . enables seamless fingerprint transplantation through parameter fusion, eliminating full-parameter updates while maintaining integrity. |
| Outcome: | The proposed framework achieves superior robustness against various scenarios while reducing computational overhead compared to traditional approaches. |
GNNavi: Navigating the Information Flow in Large Language Models by Graph Neural Network (2024.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) exhibit strong In-Context Learning (ICL) capabilities when prompts with demonstrations are used. |
| Approach: | They propose a prompt-based parameter-efficient fine-tuning approach that leverages insights into ICL’s information flow dynamics and hardwires the desired information flow into the GNN. |
| Outcome: | The proposed approach surpasses prompt-based fine-tuning methods in few-shot settings by updating just 0.2% to 0.5% of parameters. |
pFedGPT: Hierarchically Optimizing LoRA Aggregation Weights for Personalized Federated GPT Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for fine-tuning Large Language Models (LLMs) struggle with data heterogeneity and adapt shared global knowledge to individual client needs. |
| Approach: | They propose a framework that leverages Hierarchical Bayesian Optimization (HBO) for fine-grained, personalized LoRA aggregation. |
| Outcome: | The proposed framework achieves state-of-the-art (SOTA) performance on personalized FL benchmarks while introducing only minimal (approx. 4%) additional optimization overhead. |
DyLoRA: Parameter-Efficient Tuning of Pre-trained Models using Dynamic Search-Free Low-Rank Adaptation (2023.eacl-main)
Copied to clipboard
| Challenge: | Pre-training/fine-tuning of pre-training models has become more expensive and resource-hungry. |
| Approach: | They propose a low-rank adaptation technique that trains LoRA blocks for a range of ranks instead of a single rank. |
| Outcome: | The proposed method trains LoRA blocks for a range of ranks instead of a single rank . it can train dynamic search-free models with DyLoRA at least 4 to 7 times faster than LoRA . |
StyleRemix: Interpretable Authorship Obfuscation via Distillation and Perturbation of Style Elements (2024.emnlp-main)
Copied to clipboard
| Challenge: | Authorship obfuscation methods that ignore author-specific stylistic features are often too rigid and lead to degradation of fluency and grammaticality. |
| Approach: | They propose an adaptive obfuscation method that perturbs stylistic elements of text . authors release a large set of 30K high-quality, long-form texts from a diverse set of 14 authors . |
| Outcome: | The proposed method outperforms state-of-the-art methods on an array of domains on automatic and human evaluation. |
Improving Language Identification for Code-Switched Speech: The Pivotal Role of Accented English (2026.findings-eacl)
Copied to clipboard
| Challenge: | Existing models fail to identify English spoken with the accent of the matrix (dominant) language. |
| Approach: | They propose to fine tune existing LID models with accented English to improve code-switched LID . they use a metric that captures relative ranking of identified languages often overlooked by traditional metrics. |
| Outcome: | The proposed model can be fine tuned with small amounts of accented English without degrading performance on monolingual speech. |
How Much Knowledge Can You Pack into a LoRA Adapter without Harming LLM? (2025.findings-naacl)
Copied to clipboard
Sergey Pletenev, Maria Marina, Daniil Moskovskiy, Vasily Konovalov, Pavel Braslavski, Alexander Panchenko, Mikhail Salnikov
| Challenge: | Low-rank adaptation (LoRA) is a popular training technique for updating or domain-specific adaptation of Large Language Models (LLMs). |
| Approach: | They propose to use low-rank adaptation to incorporate new facts into the LLM without compromising previously learned knowledge. |
| Outcome: | The proposed approach is harmful because the model's performance declines after such fine-tuning. |
Improve Safety Training of Large Language Models with Safety-Critical Singular Vectors Localization (2025.acl-long)
Copied to clipboard
| Challenge: | Recent work on safety training with modules such as low-rank adaptation (LoRA) to resist jailbreaks shows promise, but these approaches can inadvertently degrade a model’s general utility. |
| Approach: | They propose a plug-and-play method that locates safety-critical singular vectors within the model's parameter space and a dynamic rank number determination strategy to reduce parameter overhead. |
| Outcome: | The proposed method mitigates the impact of safety training on model utility by explicitly locating and leveraging safety-critical singular vectors within the model’s parameter space. |
SLIM: Let LLM Learn More and Forget Less with Soft LoRA and Identity Mixture (2025.naacl-long)
Copied to clipboard
| Challenge: | balancing the training budget, downstream performance, and general capabilities of large language models remains a challenge in many applications. |
| Approach: | They propose a mixture of expert framework based on Soft LoRA and Identity Mixture . SLIM allows dynamic routing between LoRA adapters and identity layers . |
| Outcome: | The proposed framework reduces training cost while maintaining general capabilities . it can be open-sourced upon publication. |
RESF: Regularized-Entropy-Sensitive Fingerprinting for Black-Box Tamper Detection of Large Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for tamper detection rely on model stability, not inherently stochastic models. |
| Approach: | They propose a hypothesis-testing method for black-box tamper detection for LLMs . they propose regularized entropy-sensitive fingerprinting to enable efficient fingerprinting . |
| Outcome: | The proposed method achieves 98.80% detection accuracy under challenging conditions . it is based on a first-order surrogate for KL divergence to identify prompts most responsive to parameter perturbations. |
Completely Modular Fine-tuning for Dynamic Language Adaptation (2026.findings-eacl)
Copied to clipboard
| Challenge: | Existing studies on multilingual fine-tuning with a fixed set of languages lack dynamic adaptability to new languages. |
| Approach: | They propose a modular fine-tuning pipeline that enables dynamic language adaptation for LLMs by first training English-centric adapters for each language separately and then merging them for arbitrary-direction translation. |
| Outcome: | The proposed pipeline achieves 86% performance over traditional fine-tuning on four languages, while training only 0.1% parameters and relying on English as a bridge language without catastrophic forgetting. |
Sparse Low-rank Adaptation of Pre-trained Language Models (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for fine-tuning pre-trained large language models in a parameter-efficient manner are gaining traction within the research community. |
| Approach: | They propose a method of low-rank adaptation that enables dynamic adjustments to the intrinsic rank during the adaptation process. |
| Outcome: | The proposed approach outperforms the current method with a fixed and unalterable intrinsic rank and a low-rank adaptation process. |
BBTv2: Towards a Gradient-Free Future with Large Language Models (2022.emnlp-main)
Copied to clipboard
| Challenge: | Recent work on parameter-efficient tuning (PET) only tunes a small portion of parameters while keeping most of the parameters of the LLM unchanged. |
| Approach: | They propose an improved version of Black-Box Tuning to tune PTMs through gradient descent . they prepend continuous prompts to every layer of the PTM and propose a divide-and-conquer gradient-free algorithm to optimize the prompts alternately. |
| Outcome: | The proposed method achieves comparable performance to full model tuning and state-of-the-art parameter-efficient methods under few-shot settings while maintaining much fewer tunable parameters. |
On the Representation Geometry of LoRA Model Merging (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for low-rank Adaptation (LoRA) fine-tuning focus on globally shared structure . combining SVD with CUR improves performance of LoRA model merging . |
| Approach: | They propose a training-free method that combines SVD and CUR decomposition to improve LoRA merging performance. |
| Outcome: | The proposed procedure improves on vision and language benchmarks. |
LoraRetriever: Input-Aware LoRA Retrieval and Composition for Mixed Tasks in the Wild (2024.findings-acl)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) is an effective yet efficient solution for fine-tuning large language models. |
| Approach: | They propose a low-rank Adaptation framework that retrieves and composes multiple LoRAs according to input prompts. |
| Outcome: | Experimental results show that LoraRetriever outperforms baselines in terms of performance and versatility. |
SuLoRA: Subspace Low-Rank Adaptation for Parameter-Efficient Fine-Tuning (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for fine-tuning large language models (LLMs) introduce parameter interference, leading to a gap in generalization performance for specific tasks compared to full fine-uning. |
| Approach: | They propose a parameter-separated low-rank adapter to account for task differences by decomposing LoRA’s parameter matrix into multiple independent subspaces and assigning them differentially to distinct tasks. |
| Outcome: | The proposed method outperforms LoRA in trainable parameter efficiency and overall model performance on various NLP tasks. |
AutoLoRA: Automatically Tuning Matrix Ranks in Low-Rank Adaptation Based on Meta Learning (2024.naacl-long)
Copied to clipboard
| Challenge: | Large-scale pretraining followed by task-specific finetuning has achieved great success in various NLP tasks. |
| Approach: | They propose a meta learning based framework for automatically identifying the optimal rank of each LoRA layer. |
| Outcome: | The proposed framework is based on a meta learning based framework that can identify the optimal rank of each LoRA layer. |
Federated LoRA Fine-Tuning with Pipelined Error-Mitigated Aggregation and Matrix-Wise Freezing (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for fine-tuning large language models often suffer from biased model aggregation and are hindered by significant communication and computation burden. |
| Approach: | They propose a Federated low-rank adaptation system for large language models that leverages pipelined error-mitigated model aggregation and adaptive matrix-wise parameter freezing to mitigate aggregations. |
| Outcome: | The proposed system improves time-to-target by 2.17-8.48 on real-world datasets. |
MoLA: MoE LoRA with Layer-wise Expert Allocation (2025.findings-naacl)
Copied to clipboard
Chongyang Gao, Kezhen Chen, Jinmeng Rao, Ruibo Liu, Baochen Sun, Yawen Zhang, Daiyi Peng, Xiaoyuan Guo, Vs Subrahmanian
| Challenge: | Recent efforts to integrate low-rank adaptation (LoRA) with the Mixture-of-Experts (MoE) have achieved performance comparable to full-parameter fine-tuning by tuning much fewer parameters. |
| Approach: | They propose a parameter-efficient MoE method for low-rank adaptation with the Mixture-of-Experts (MoE) they use layers of LoRA experts to allocate more LoRA expert to middle layers . |
| Outcome: | The proposed method outperforms baseline models on six well-known NLP and commonsense QA benchmarks on LLAMA-2, Mistral, and Gemma. |
Enhancing Multimodal Continual Instruction Tuning with BranchLoRA (2025.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to fine tune Multimodal Large Language Models (MLLMs) are prone to Catastrophic Forgetting (CF) existing approaches rely on the Mixture-of-Experts (MoE) LoRA framework to preserve previous instruction alignments. |
| Approach: | They propose an asymmetric tuning-freezing mechanism to mitigate parameter inefficiency . branch-specific routers are introduced to ensure optimal branch distribution over time . |
| Outcome: | The proposed framework outperforms existing frameworks on the latest MCIT benchmarks. |
Beyond Demonstrations: Dynamic Vector Construction from Latent Representations (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for In-Context Learning (ICL) are sensitive to ICL-specific factors and rely on heuristic-based injection positions. |
| Approach: | They propose a method that extracts task-relevant representations from large language models and reinjects them during inference. |
| Outcome: | The proposed method outperforms few-shot In-Context Learning (ICL) and LoRA methods without repeated demonstration processing. |
VoiceTextBlender: Augmenting Large Language Models with Speech Capabilities via Single-Stage Joint Speech-Text Supervised Fine-Tuning (2025.naacl-long)
Copied to clipboard
Yifan Peng, Krishna C Puvvada, Zhehuai Chen, Piotr Zelasko, He Huang, Kunal Dhawan, Ke Hu, Shinji Watanabe, Jagadeesh Balam, Boris Ginsburg
| Challenge: | Recent studies have augmented large language models (LLMs) with speech capabilities, leading to the development of speech language models. |
| Approach: | They propose a single-stage joint speech-text SFT approach for training SpeechLMs . their model combines text-only SFT data with three types of speech-related data . |
| Outcome: | The proposed model outperforms previous SpeechLMs on speech-based QA tasks while maintaining original speech-only capabilities. |
DreamSync: Aligning Text-to-Image Generation with Image Understanding Feedback (2025.naacl-long)
Copied to clipboard
Jiao Sun, Deqing Fu, Yushi Hu, Su Wang, Royi Rassin, Da-Cheng Juan, Dana Alon, Charles Herrmann, Sjoerd Van Steenkiste, Ranjay Krishna, Cyrus Rashtchian
| Challenge: | Text-to-Image models (T2I) still struggle to produce images that are both aesthetically pleasing and faithful to the user’s input text. |
| Approach: | They propose a training algorithm that trains T2I models to be faithful to the input text. |
| Outcome: | The proposed model improves both the semantic alignment and aesthetic appeal of two diffusion-based T2I models, evidenced by multiple benchmarks (+1.7% on TIFA, +2.9% on DSG1K, +3.4% on VILA aesthetic). |
DeAR: Dual-Stage Document Reranking with Reasoning Agents via LLM Distillation (2025.findings-emnlp)
Copied to clipboard
| Challenge: | DeAR is an open-source framework that decouples the tasks of LLMs with holistic cross-document analysis. |
| Approach: | They propose an open-source framework that decouples relevance scoring with holistic cross-document analysis. |
| Outcome: | The proposed framework outperforms open-source frameworks in QA and open-domain QA. |
Fast Randomized Low-Rank Adaptation of Pre-trained Language Models with PAC Regularization (2024.findings-acl)
Copied to clipboard
| Challenge: | Low-rank adaptation (LoRA) is an efficient way to fine-tune large language models (LLMs) but its memory overhead restricts it to scale up when the model size increases. |
| Approach: | They propose a low-rank adaptation method which decomposes model weight updates into a pair of low-ranked projection matrices and incorporates a regularizer to improve generalization. |
| Outcome: | The proposed method achieves better performance under few-shot settings and similar performance to the SOTA low-rank adaptation methods. |
Forgetting before Learning: Utilizing Parametric Arithmetic for Knowledge Updating in Large Language Models (2024.acl-long)
Copied to clipboard
| Challenge: | Existing methods of model editing and knowledge updating add additional network parameters, knowledge bases, knowledge base, and model parameters. |
| Approach: | They propose a new paradigm for fine-tuning called F-Learning that employs parametric arithmetic to facilitate the forgetting of old knowledge and learning of new knowledge. |
| Outcome: | The proposed model outperforms existing models on two datasets and is comparable to full fine-tuning and LoRA fine-uning. |
A Deep Dive into the Trade-Offs of Parameter-Efficient Preference Alignment Techniques (2024.acl-long)
Copied to clipboard
| Challenge: | Large language models are pre-trained on trillions of tokens and instruction-tuned or aligned to specific preferences. |
| Approach: | They propose guidelines to help researchers perform more effective parameter-efficient LLM alignment. |
| Outcome: | The proposed methods outperform preference optimization and outperformed pre-trained models on three key axes. |
Zero-Shot Cross-Domain Dialogue State Tracking via Dual Low-Rank Adaptation (2024.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to zero-shot dialogue state tracking (DST) involve embedding prompts into language models, but these methods have inherent limitations. |
| Approach: | They propose a plug-and-play architecture designed for zero-shot dialogue state tracking (DST) dual low-rank adaptation targets dialogue context processing and prompt optimization without incurring additional inference latency. |
| Outcome: | The proposed architecture outperforms baseline methods on multi-domain datasets and the MultiWOZ dataset. |
MALoRA: Mixture of Asymmetric Low-Rank Adaptation for Enhanced Multi-Task Learning (2025.findings-naacl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) can be fine-tuned to new tasks, but in multi-task scenarios, training imbalance and seesaw effect often arise. |
| Approach: | They propose a flexible fine-tuning framework that leverages asymmetric optimization among LoRA experts to reduce training imbalance and improve performance. |
| Outcome: | The proposed framework outperforms baseline methods in inter- and intra-task learning scenarios. |
Event Detection with a Context-Aware Encoder and LoRA for Improved Performance on Long-Tailed Classes (2026.findings-eacl)
Copied to clipboard
| Challenge: | Decoder-only models dominate the event detection literature, but their unidirectional attention mechanism has been a roadblock in getting strong performance on embedding. |
| Approach: | They propose to use Macro-F1 as a more representative measure of a model’s ability across the long-tail of event types to improve their models' performance. |
| Outcome: | The proposed model improves on the decoder-only models, showing that low-rank Adaptation can be an effective tool to enhance LLMs’ performance on long-tailed event classes. |
Role-Conditioned Refusals: Evaluating Access Control Reasoning in Large Language Models (2026.findings-eacl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) blur role boundaries by producing unrestricted responses. |
| Approach: | They propose to extend the Spider and BIRD text-to-SQL datasets with real-time PostgreSQl role-based policies at the table and column levels. |
| Outcome: | The proposed model improves refusal precision and lowers false permits. |
CoVoGER: A Multilingual Multitask Benchmark for Speech-to-text Generative Error Correction with Large Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language models can fix recognition or translation errors that traditional rescoring cannot fix. |
| Approach: | They propose a benchmark for GER that covers both ASR and speech-to-text translation across 15 languages and 28 language pairs. |
| Outcome: | The proposed benchmark is built on common voice 20.0 and CoVoST-2 with Whisper and SeamlessM4T. |
Tiny Budgets, Big Gains: Parameter Placement Strategy in Parameter Super-Efficient Fine-Tuning (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods such as LoRA and VeRA use memory-efficient methods to fine-tune large language models. |
| Approach: | They propose a method that uses only 1–5% of the standard LoRA parameters and achieves state-of-the-art performance across a wide range of tasks. |
| Outcome: | The proposed method achieves state-of-the-art performance across a wide range of tasks using only 1–5% of the standard LoRA parameters. |
RankAdaptor: Hierarchical Rank Allocation for Efficient Fine-Tuning Pruned LLMs via Performance Model (2025.findings-naacl)
Copied to clipboard
| Challenge: | Current compression techniques entail structural pruning and a recovery phase that leverages the Low-Rank Adaptation algorithm. |
| Approach: | They propose a hierarchical rank allocation method that enables efficient fine-tuning of pruned LLMs according to layerwise specific recovery requirements. |
| Outcome: | The proposed algorithm outperforms state-of-the-art methods across pruning settings and LLM architectures with improvements ranging from 0.7% to 5.5%. |
On Mitigating Performance Disparities in Multilingual Speech Recognition (2024.emnlp-main)
Copied to clipboard
| Challenge: | Automatic Speech Recognition systems are not always equally effective for all users, and gender disparity in their performance is a significant concern. |
| Approach: | They compare performance of different fine-tuning algorithms for multilingual speech recognition across languages and genders. |
| Outcome: | The proposed algorithms improve performance and parity across languages and languages. |
Instance-Level Dynamic LoRAs Composition for Cross-Task Generalization (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Large language models perform well on tasks that have undergone fine-tuning of instructions, but performance on completely unseen tasks is often less than ideal. |
| Approach: | They propose a task-level LoRAs combination which learns the LoRA modules combination weights based on a small number of samples to form the task model. |
| Outcome: | The proposed method outperforms the typical method, LoraHub, on 16 out of 27 tasks. |
Towards More Efficient Post-training via Fourier Domain Adapter Framework (2025.findings-emnlp)
Copied to clipboard
| Challenge: | FDA reparameterizes the core projection operation of the adapter module directly in the Fourier domain. |
| Approach: | They propose a framework that reparameterizes the core projection operation of the adapter module directly in the Fourier domain. |
| Outcome: | The proposed framework outperforms existing parameter-efficient fine-tuning methods on GLUE, E2E NLG, and instruction tuning benchmarks. |
HD-PiSSA: High-Rank Distributed Orthogonal Adaptation (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for large language models constrain update to low-rank subspaces, limiting expressiveness and performance. |
| Approach: | They propose a distributed PEFT approach that initializes adapters across different devices and aggregates their delta updates collectively on (W) Empirically, HD-PiSSA provides 16 higher effective updated ranks than data-parallel LoRA or PiSSA when fine-tuning on 8 GPUs with the same per-device adapter rank. |
| Outcome: | Empirically, HD-PiSSA outperforms LoRA and PiSSA in math, code, and multi-task learning tasks. |
Spectral Disentanglement: Rank-Aware Task Adaptation for Rehearsal-free Continual Learning in LLMs (2026.acl-long)
Copied to clipboard
| Challenge: | Continual Learning (CL) for Large Language Models faces a fundamental Stability-Plasticity Dilemma . Rank-Blindness enforces a single rank constraint across diverse tasks, leading to catastrophic forgetting of earlier tasks and underfitting on complex new ones. |
| Approach: | They propose a rank-spectrum-based rehearsal-free framework that explicitly disentangles knowledge into two orthogonal subspaces. |
| Outcome: | The proposed framework achieves a superior stability-plasticity balance compared to single-rank baselines. |
AGGC: Adaptive Group Gradient Clipping for Stabilizing Large Language Model Training (2026.findings-acl)
Copied to clipboard
| Challenge: | Adaptive group-wise gradient clipping (AGGC) is a new approach to stabilize training of Large Language Models. |
| Approach: | They propose a method to stabilize gradient clipping by partitioning parameters into groups based on functional types and a time-dependent scheduling mechanism to balance exploration and convergence. |
| Outcome: | The proposed algorithm outperforms standard LoRA and achieves 72.93% accuracy . it can be integrated into existing pipelines with negligible overhead. |
LoSiA: Efficient High-Rank Fine-Tuning via Subnet Localization and Optimization (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods perform extensive matrix multiplications in domain specialization tasks, resulting in computational inefficiency and sub-optimal fine-tuning performance. |
| Approach: | They propose a method that localizes and optimizes critical parameters during training . they propose 'LoSiA-Pro' which reduces training latency by 27% . |
| Outcome: | The proposed method achieves minimal performance drop compared to full fine-tuning while requiring the least training time across domain specialization and common-sense reasoning tasks. |
LEO-MINI: An Efficient Multimodal Large Language Model using Conditional Token Reduction and Mixture of Multi-Modal Experts (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent approaches to reduce visual tokens have been criticized for their computational efficiency and lack of visual reasoning capabilities. |
| Approach: | They propose a novel multi-modal large language model that reduces the number of visual tokens and simultaneously boosts visual reasoning capabilities. |
| Outcome: | The proposed model significantly reduces the number of visual tokens and boosts visual reasoning capabilities. |
LoRA-drop: Efficient LoRA Parameter Pruning based on Output Evaluation (2025.coling-main)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) is currently the most commonly used PEFT method for fine-tuning models with billions of parameters. |
| Approach: | They propose to use low-rank Adaptation to evaluate LoRA parameter features and then retain LoRA for important layers and the other layers share the same LoRA. |
| Outcome: | The proposed method achieves comparable performance to full fine-tuning and LoRA while retaining 50% of the LoRA parameters on average. |
AdaMix: Mixture-of-Adaptations for Parameter-efficient Model Tuning (2022.emnlp-main)
Copied to clipboard
Yaqing Wang, Sahaj Agarwal, Subhabrata Mukherjee, Xiaodong Liu, Jing Gao, Ahmed Hassan Awadallah, Jianfeng Gao
| Challenge: | Standard fine-tuning of large pre-trained language models requires updating hundreds of millions to billions of parameters and storing a large copy of the PLM weights for every task. |
| Approach: | They propose a parameter-efficient fine-tuning technique where small trainable components are injected into the PLM and updated during fine-uning. |
| Outcome: | The proposed method outperforms SOTA parameter-efficient fine-tuning and full model fine-uning on GLUE development set with RoBERTa-large encoder. |
Structuring Radiology Reports: Challenging LLMs with Lightweight Models (2025.emnlp-main)
Copied to clipboard
Johannes Moll, Louisa Fay, Asfandyar Azhar, Sophie Ostmeier, Sergios Gatidis, Tim C. Lueth, Curtis Langlotz, Jean-Benoit Delbrouck
| Challenge: | Radiology reports lack a standardized format, limiting both interpretability and machine learning applications. |
| Approach: | They propose to use lightweight encoder-decoder models for structuring radiology reports . they compare models with eight open-source LLMs with prompting and in-context learning . |
| Outcome: | The proposed models outperform eight open-source LLMs on a human-annotated test set. |
LLM Braces: Straightening Out LLM Predictions with Relevant Sub-Updates (2025.acl-long)
Copied to clipboard
| Challenge: | Recent studies reveal that much of the knowledge in a Transformer-based Large Language Model (LLM) is encoded in its feed-forward (FFN) layers, where each FNN layer can be interpreted as the summation of sub-updates, each corresponding to a weighted column vector from the FFN’s value parameter matrix. |
| Approach: | They propose a method that computes relevance scores associated with value vectors in FFN layers and leverages these scores to dynamically adjust the contribution of sub-updates. |
| Outcome: | The proposed framework outperforms baseline approaches in fine-tuning and zero-shot settings while requiring significantly fewer tunable parameters. |
KnowLA: Enhancing Parameter-efficient Finetuning with Knowledgeable Adaptation (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing methods for parameter-efficient finetuning (PEFT) are limited and only finetune a small number of parameters using limited instruction data. |
| Approach: | They propose a method that inserts an adaptation layer into an LLM to integrate embeddings of entities appearing in the input text. |
| Outcome: | The proposed method can activate parameterized knowledge in an LLM without changing its parameters or input prompts. |
GuiLoMo: Allocating Experts and Ranks for LoRA-MoE via Bilevel Optimization with GuidedSelection Vectors (2025.findings-emnlp)
Copied to clipboard
Xinrong Chen, Hengyuan Zhang, Yingmin Qiu, Xiao Liang, Ziyue Li, Guanyu Wang, Weiping Li, Tong Mo, Hayden Kwok-Hay So, Ngai Wong
| Challenge: | Low-Rank Adaptation (LoRA) methods are efficient for a large language model with reduced computational costs. |
| Approach: | They propose a layer-wise expert numbers and ranks allocation strategy with GuidedSelection Vectors. |
| Outcome: | The proposed method achieves superior or comparable performance to all baselines on three backbone models. |
UNLEARN Efficient Removal of Knowledge in Large Language Models (2025.findings-naacl)
Copied to clipboard
| Challenge: | Large Language Models excel in many tasks but are outperformed by specialized tools for certain tasks. |
| Approach: | They propose a method that uses subspace techniques to selectively remove knowledge . they propose 'unlearn' method that can forget or unlear the knowledge without retraining . |
| Outcome: | The proposed method outperforms existing methods for forgetting target knowledge while preserving related knowledge. |
Parameter-Efficient Instruction Tuning of Large Language Models For Extreme Financial Numeral Labelling (2024.naacl-long)
Copied to clipboard
Subhendu Khatuya, Rajdeep Mukherjee, Akash Ghosh, Manjunath Hegde, Koustuv Dasgupta, Niloy Ganguly, Saptarshi Ghosh, Pawan Goyal
| Challenge: | Existing methods to automatically annotate relevant numerals (GAAP metrics) occurring in financial documents are not cost-effective nor scalable. |
| Approach: | They propose a generative paradigm for annotating GAAP metrics with XBRL tags using metric metadata and a parameter efficient model using LoRA. |
| Outcome: | The proposed model outperforms baseline models on two financial numeric labeling datasets and outperformed several strong baseline models. |
One Network, Many Masks: Towards More Parameter-Efficient Transfer Learning (2023.acl-long)
Copied to clipboard
| Challenge: | Parameter-efficient transfer learning methods can be expensive in storage when applied to broader ranges of tasks. |
| Approach: | They propose a method that enables efficient sharing of a single PETL network across layers and tasks. |
| Outcome: | The proposed method outperforms other methods with 10% parameters required by the latter on various downstream tasks. |
DMoERM: Recipes of Mixture-of-Experts for Effective Reward Modeling (2024.findings-acl)
Copied to clipboard
| Challenge: | Using a reward model (RM) to improve the effectiveness of large language models, there are two challenges in training. |
| Approach: | They propose a reward model (RM) that is a proxy of human preferences and assigns scores to the outputs of the large language model (LLM) a human annotation consistency rate of 60% to 75% is causing training data to contain a lot of noise. |
| Outcome: | The proposed model outperforms state-of-the-art ensemble methods and mitigates the overoptimization problem. |
X-FLoRA: Cross-modal Federated Learning with Modality-expert LoRA for Medical VQA (2025.emnlp-main)
Copied to clipboard
| Challenge: | Medical visual question answering (VQA) and federated learning (FL) are important tools for privacy-preserving collaborative learning. |
| Approach: | They propose a cross-modal FL framework that uses modality-expert low-rank adaptation for medical visual question answering (VQA) X-FLoRA enables the synthesis of images from one modality to another without requiring data sharing . |
| Outcome: | Experiments show that X-FLoRA outperforms existing FL methods in terms of performance . XFLorage enables synthesis of images from one modality to another without data sharing . |
More Than Catastrophic Forgetting: Integrating General Capabilities For Domain-Specific LLMs (2024.emnlp-main)
Copied to clipboard
Chengyuan Liu, Yangyang Kang, Shihang Wang, Lizhi Qing, Fubang Zhao, Chao Wu, Changlong Sun, Kun Kuang, Fei Wu
| Challenge: | a recent study shows that performance on general tasks decreases after Large Language Models are fine-tuned on domain-specific tasks. |
| Approach: | They propose a general capability integration approach to integrate general capabilities and domain knowledge within a single instance. |
| Outcome: | The proposed method improves performance on domain-specific tasks by integrating general capabilities and domain knowledge. |
RoLoRA: Fine-tuning Rotated Outlier-free LLMs for Effective Weight-Activation Quantization (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) improves training efficiency by updating only a small portion of the weights in Large Language Models. |
| Approach: | They propose a rotation-aware scheme to fine-tune rotated outlier-free LLMs for effective weight-activation quantization. |
| Outcome: | The proposed method improves low-bit LoRA convergence and post-training quantization robustness. |
Mixture-of-Subspaces in Low-Rank Adaptation (2024.emnlp-main)
Copied to clipboard
| Challenge: | Using a subspace-inspired Low-Rank Adaptation method, large language models can be optimized for downstream tasks using parameter-efficient finetuning. |
| Approach: | They propose a subspace-inspired Low-Rank Adaptation method that decomposes LoRA weights into two subspaces and merges them into the frozen original weight. |
| Outcome: | The proposed method outperforms LoRA on commonsense reasoning, visual instruction tuning, and subject-driven text-to-image generation tasks. |
A Layer-wise Analysis of Supervised Fine-Tuning (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for fine-tuning ignore depth-dependent heterogeneity of instruction-following . a critical gap remains in understanding where these changes occur across the model's depth and which layers are essential for instruction- following. |
| Approach: | They propose a method which selectively updates critical intermediate layers . they show that effective alignment is architecturally localized rather than distributed . |
| Outcome: | The proposed method outperforms standard LoRA up to 10.2% on GSM8K with reduced parameter overhead. |
DistillMIKE: Editing Distillation of Massive In-Context Knowledge Editing in Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | In-context knowledge editing has shown respectable abilities on knowledge editing in terms of generalization and specificity. |
| Approach: | They propose a novel extension of in-context knowledge editing (IKE) that allows for massive edits to be injected into large language models. |
| Outcome: | The proposed method shows state-of-the-art perfomrances and comparable performance with MIKE. |
MoDE: Effective Multi-task Parameter Efficient Fine-Tuning with a Mixture of Dyadic Experts (2025.findings-naacl)
Copied to clipboard
| Challenge: | Recent efforts have explored mixtures of LoRA modules for multi-task settings, but this study reveals redundancy in the down-projection matrix of these architectures. |
| Approach: | They propose a method to share down-projection matrix across tasks and employ atomic rank-one adapters coupled with routers that allow more sophisticated task-level specialization. |
| Outcome: | The proposed method outperforms state-of-the-art models on a SNI benchmark and provides a practical solution for deploying lightweight models. |
AdaDHP: Fine-Grained Fine-Tuning via Dual Hadamard Product and Adaptive Parameter Selection (2025.acl-long)
Copied to clipboard
| Challenge: | Increasing number of parameters can be challenging under resource-constrained environments. |
| Approach: | They propose a parameter-efficient fine-tuning method with fewer parameters and finer granularity that can adaptively select important parameters for each task. |
| Outcome: | The proposed method can fine-tune important parameters for each task, while maintaining the same weights. |
LaRA: Large Rank Adaptation for Speech and Text Cross-Modal Learning in Large Language Models (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to integrate speech and text capabilities into large language models (LLMs) require significantly larger ranks comparable to the pretrained weights to accommodate the complexities of speech-text cross-modality learning. |
| Approach: | They propose a large-rank adaptive approach for cross-modal integration of speech and text into large language models (LLMs) it uses a Hi-Fi vocoder to synthesize speech waveforms from the generated speech units. |
| Outcome: | The proposed model can be extended to other cross-modal applications. |
Tied-LoRA: Enhancing parameter efficiency of LoRA with Weight Tying (2024.naacl-long)
Copied to clipboard
| Challenge: | a new paradigm for low-rank Adaptation (LoRA) uses weight tying and selective training to improve parameter efficiency. |
| Approach: | They propose a paradigm that uses weight tying and selective training to enhance parameter efficiency of Low-rank Adaptation. |
| Outcome: | The proposed paradigm achieves comparable performance to LoRA with reduced model complexity . the proposed paradigm can be used for a variety of tasks and languages . |
Meta-Tool: Efficient Few-Shot Tool Adaptation for Small Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Using a Llama-3.2-3B-Instruct backbone, we evaluate four adaptation mechanisms across four benchmarks: Gorilla APIBench, Spider 2.0, WebArena, and InterCode. |
| Approach: | They compare hypernetwork-based LoRA adaptation against carefully designed few-shot prompting in a controlled experiment . they find that few- shot prompting contributes +21.5% to performance and documentation contributes 0% . |
| Outcome: | The hypernetwork-based LoRA adaptation provides no measurable improvement over few-shot prompting alone. |
DenseLoRA: Dense Low-Rank Adaptation of Large Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Low-rank adaptation (LoRA) is an efficient approach for adapting large language models (LLMs) but many of the weights in these matrices are redundant, leading to inefficiencies in parameter utilization. |
| Approach: | They propose a low-rank adaptation approach that fine-tunes two low-ranked matrices and adapts them through a dense low-Rank matrix, improving parameter utilization and adaptation efficiency. |
| Outcome: | The proposed approach achieves 83.8% accuracy with only 0.01% of trainable parameters compared to LoRA's 80.8% with 0.70% of trainability parameters on LLaMA3-8B. |
Propulsion: Steering LLM with Tiny Fine-Tuning (2025.coling-main)
Copied to clipboard
| Challenge: | Propulsion is a parameter-efficient fine-tuning method that selectively re-scales specific dimensions of a pre-trained model without modifying the model’s parameters. |
| Approach: | They propose a parameter-efficient fine-tuning method that selectively re-scales specific dimensions of a pre-trained model without modifying the parameters. |
| Outcome: | The proposed method reduces parameter count from 355.3 million to 0.086 million while maintaining competitive performance across benchmarks. |
SSH: Sparse Spectrum Adaptation via Discrete Hartley Transformation (2025.naacl-long)
Copied to clipboard
| Challenge: | Low-rank adaptation (LoRA) has been demonstrated effective in reducing the trainable parameter number when fine-tuning a large foundation model (LLM). |
| Approach: | They propose a low-rank adaptation approach that reduces the number of trainable parameters while enhancing model performance. |
| Outcome: | The proposed approach outperforms existing parameter-efficient fine-tuning methods while achieving substantial reductions in computational cost and memory requirements. |
ResLoRA: Identity Residual Mapping in Low-Rank Adaption (2024.findings-acl)
Copied to clipboard
Shuhua Shi, Shaohan Huang, Minghui Song, Zhoujun Li, Zihan Zhang, Haizhen Huang, Furu Wei, Weiwei Deng, Feng Sun, Qi Zhang
| Challenge: | Low-rank adaptation (LoRA) is one of the most popular parameter-efficient fine-tuning methods. |
| Approach: | They propose a low-rank adaptation method that adds residual paths during training and merges them together during inference to achieve better results. |
| Outcome: | The proposed method achieves 2.5x faster convergence speed and improves performance by 14.3% on NLG, NLU, and text-to-image tasks. |
LoRMA: Low-Rank Multiplicative Adaptation for LLMs (2025.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models have shown impressive generalization capabilities, but can be expensive to fine-tune due to high computational costs. |
| Approach: | They propose a low-rank multiplicative Adaptation technique that shifts the paradigm of additive updates to a richer space of matrix multiplicative transformations. |
| Outcome: | The proposed approach overcomes computational complexity and rank bottlenecks in terms of matrix multiplication metrics. |
Fast Forwarding Low-Rank Training (2024.emnlp-main)
Copied to clipboard
| Challenge: | Modern optimizers provide a spectacular array of tweaks to stabilize training trajectories and accelerate Stochastic Gradient Descent (SGD). |
| Approach: | They propose a fast-forward approach to accelerate large segments of SGD training . they alternate between Adam SGD for burn-in and accelerating by line search . |
| Outcome: | The proposed approach speeds up training without compromising model performance. |
Dodo: Dynamic Contextual Compression for Decoder-only LMs (2024.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to NLP are sparsifying attention patterns or approximating the attention computation with kernel methods. |
| Approach: | They propose a method for dynamic contextual compression for decoder-only LMs. |
| Outcome: | The proposed method reduces the cost of self-attention to a fraction of typical time and space. |
Beware of Your Po! Measuring and Mitigating AI Safety Risks in Role-Play Fine-Tuning of LLMs (2025.acl-long)
Copied to clipboard
Weixiang Zhao, Yulin Hu, Yang Deng, Jiahe Guo, Xingyu Sui, Xinyang Han, An Zhang, Yanyan Zhao, Bing Qin, Tat-Seng Chua, Ting Liu
| Challenge: | Existing role-play fine-tuning techniques improve role adaptability but may degrade safety performance, especially for villainous characters. |
| Approach: | They propose safety-aware Role-Play Fine-Tuning (SaRFT) to balance role-playing capabilities and safety. |
| Outcome: | The proposed method outperforms state-of-the-art baselines under both LoRA and full-parameter fine-tuning settings. |
NeuroAda: Activating Each Neuron’s Potential for Parameter-Efficient Fine-Tuning (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for parameter-efficient fine-tuning are limited and require computational and memory resources. |
| Approach: | They propose a parameter-efficient fine-tuning method that enables fine-grained model finetunation while maintaining high memory efficiency. |
| Outcome: | The proposed method reduces CUDA memory usage by up to 60% while maintaining high performance. |
Capability Decomposition for Unified Information Extraction via Hierarchical Mixture-of-Experts (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for IE tasks suffer from inconsistent schema representation and implicitly intermediate reasoning . UC-UIE adopts a low-rank adapted hierarchical Mixture-of-Experts adapter for UIE tasks . |
| Approach: | They propose a framework that decomposes IE reasoning into three universal capabilities . UC-UIE adopts a low-rank Adaptation adapter to fine-tune LLMs for IE tasks . |
| Outcome: | The proposed framework outperforms full-parameter tuning methods with 1.24% trainable parameters and outperformed existing methods in generalization and interpretability. |
Measuring What Matters: Evaluating Ensemble LLMs with Label Refinement in Inductive Coding (2025.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) are prone to inconsistencies and individual biases, limiting their reliability. |
| Approach: | They propose a framework that combines ensemble methods with code refinement methodology to address these challenges. |
| Outcome: | The proposed framework outperforms large language models and LLMs with a low-rank averaging and a moderator-based mechanism to simulate human consensus. |
Towards Infinite-Long Prefix in Transformer (2025.emnlp-main)
Copied to clipboard
| Challenge: | Prefix Learning is an empirically efficient and effective method for language models . but the theoretical understandings are limited on the performance of such methods . |
| Approach: | They propose a method that can train an ultra-long prefix in a stylized setting using the Neural Tangent Kernel framework. |
| Outcome: | The proposed method can achieve superior performance on vision, natural language, and math data. |
LaMDA: Large Model Fine-Tuning via Spectrally Decomposed Low-Dimensional Adaptation (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Low-rank adaptation (LoRA) fine-tunes large language models due to its significant reduction in trainable parameters, but its backward updates require storing high-dimensional intermediate activations and optimizer states, requiring high peak GPU memory. |
| Approach: | They propose a low-dimensional adaptation approach to fine-tune large language models which freezes a first projection matrix while introducing a lower-dimensional trainable square matrix. |
| Outcome: | The proposed approach reduces trainable parameters and peak GPU memory footprint while preserving low-dimensional trainable square matrix. |
WeightLoRA: Keep Only Necessary Adapters (2026.acl-long)
Copied to clipboard
| Challenge: | Low-rank adaptation (LoRA) adds trainable adapters to selected layers, but requires significant memory to train large models and intuition on which layers to add adapters. |
| Approach: | They propose a method which adds trainable adapters to selected layers . they compare weightLoRA with different adaptive approaches to reduce trainable parameters while maintaining consistent or even superior metric values. |
| Outcome: | The proposed method reduces the number of trainable parameters while maintaining the capability to obtain consistent or even superior metric values. |
META-LORA: Memory-Efficient Sample Reweighting for Fine-Tuning Large Language Models (2025.coling-main)
Copied to clipboard
| Challenge: | Supervised fine-tuning (SFT) is widely adopted for tailoring large language models (LLMs) to specific downstream tasks. |
| Approach: | They propose a memory-efficient method for automatic sample reweighting that learns to re-weight fine-tuning samples by minimizing the loss on a small, high-quality validation set. |
| Outcome: | Meta-LoRA learns to reweight fine-tuning samples by minimizing the loss on a small, high-quality validation set through an end-to-end bi-level optimization framework based on meta-learning. |
UORA: Uniform Orthogonal Reinitialization Adaptation in Parameter Efficient Fine-Tuning of Large Models (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods such as LoRA and VeRA use a low-rank approximation method that reduces the number of trainable parameters without compromising performance. |
| Approach: | They propose a parameter-efficient fine-tuning approach that leverages a low-rank approximation method that reduces the number of trainable parameters without compromising performance. |
| Outcome: | The proposed approach outperforms existing methods on GLUE and E2E benchmarks and is effective in instruction-tuning large language models and image classification models. |
BeamLoRA: Beam-Constraint Low-Rank Adaptation (2025.acl-long)
Copied to clipboard
Naibin Gu, Zhenyu Zhang, Xiyu Liu, Peng Fu, Zheng Lin, Shuohuan Wang, Yu Sun, Hua Wu, Weiping Wang, Haifeng Wang
| Challenge: | Low-Rank Adaptation (LoRA) is one of the most efficient parameter-efficient fine-tuning methods. |
| Approach: | They propose to conceptualize each LoRA module as a beam where each rank corresponds to a potential sub-solution. |
| Outcome: | The proposed method improves performance on three base models and 12 datasets. |
Fisher Information-based Efficient Curriculum Federated Learning with Large Language Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing frameworks for learning Large Language Models (LLMs) require adaptive data processing and low-rank adjustment to improve accuracy and fine-tuning speed. |
| Approach: | They propose a fisher information-based adaptive federated curriculum learning framework with two novel methods to improve FL fine-tuning process. |
| Outcome: | The proposed framework improves performance and fine-tuning speed compared with baseline approaches. |
PMSS: Pretrained Matrices Skeleton Selection for LLM Fine-tuning (2025.coling-main)
Copied to clipboard
| Challenge: | Low-rank adaptation and its variants have been popular due to their ability to avoid excessive inference costs. |
| Approach: | They propose a low-rank adaptation method that enables high-rank updates with low costs while leveraging semantic and linguistic information inherent in pre-trained weight. |
| Outcome: | The proposed method outperforms LoRA and other fine-tuning methods across tasks with less trainable parameters. |
Fair Federated Learning with Biased Vision-Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing literature ignores the inherent group unfairness within CLIP and its ethical implications on FL applications. |
| Approach: | They propose a fairness-aware adaptation framework for CLIP in federated learning . they propose to leverage biased pre-trained VLMs to build fair FL frameworks . |
| Outcome: | The proposed framework addresses unique bias in FL, triggered by data heterogeneity . it trains a fair FL model with fairness-aware deep visual prompting (DVP) Extensive results on human face attribute recognition (FAR) applications show it outperforms state-of-the-art FL models . |
Promoting Data and Model Privacy in Federated Learning through Quantized LoRA (2024.findings-emnlp)
Copied to clipboard
Zhu JianHao, Changze Lv, Xiaohua Wang, Muling Wu, Wenhao Liu, Tianlong Li, Zixuan Ling, Cenyuan Zhang, Xiaoqing Zheng, Xuanjing Huang
| Challenge: | Existing federated learning frameworks require substantial data and computational resources to develop large language models. |
| Approach: | They propose a method that distributes a quantized version of the model’s parameters during training and combine it with a popular fine-tuning method to significantly reduce communication costs. |
| Outcome: | The proposed method enables accurate estimations for parameter updates while preventing clients from accessing a model whose performance is comparable to the centrally hosted one. |
Rethinking Parameter Sharing for LLM Fine-Tuning with Multiple LoRAs (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for large language models use parameter-efficient techniques such as Low-Rank Adaptation (LoRA) prior studies suggest that the inner A matrices are highly similar during training and therefore suitable for sharing. |
| Approach: | They propose an asymmetric multi-LoRA design with multiple A matrices and a single shared B in multi-task fine-tuning. |
| Outcome: | The proposed methods achieve more balanced performance across tasks with comparable or superior average accuracy relative to existing methods. |
DoRA: Enhancing Parameter-Efficient Fine-Tuning with Dynamic Rank Distribution (2024.acl-long)
Copied to clipboard
| Challenge: | Existing parameter-efficient fine-tuning methods such as Low-Rank Adaptation ignore the differential parameter budget requirements across weight matrices, which may lead to suboptimal fine-uning outcomes. |
| Approach: | They propose a parameter-efficient low-rank Adaptation method that decomposes high-rank LoRA layers into structured single-rank components and allows dynamic pruning of parameter budget . |
| Outcome: | The proposed method outperforms LoRA and LoRA with the same parameter budget and performance. |
Polynomial Expansion Rank Adaptation: Enhancing Low-Rank Fine-Tuning with High-Order Interactions (2026.findings-acl)
Copied to clipboard
| Challenge: | Low-rank adaptation (LoRA) is a widely used strategy for efficient fine-tuning of large language models, but its strictly linear structure limits expressive capacity. |
| Approach: | They propose a method that introduces structured polynomial expansion directly into the low-rank factor space. |
| Outcome: | The proposed method outperforms state-of-the-art methods across diverse benchmarks. |
CoPL: Collaborative Preference Learning for Personalizing LLMs (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for personalizing large language models struggle with flexibility and generalization. |
| Approach: | They propose a graph-based collaborative filtering framework that models user-response relationships to enhance preference estimation in sparse annotation settings. |
| Outcome: | The proposed framework outperforms existing reward models in TL;DR, UltraFeedback-P, and PersonalLLM datasets. |
TeamLoRA: Boosting Low-Rank Adaptation with Expert Collaboration and Competition (2025.acl-long)
Copied to clipboard
Tianwei Lin, Jiang Liu, Wenqiao Zhang, Yang Dai, Haoyuan Li, Zhelun Yu, Wanggui He, Juncheng Li, Jiannan Guo, Hao Jiang, Siliang Tang, Yueting Zhuang
| Challenge: | Existing methods for fine-tuning are resource-efficient, but performance often falls short . a new approach, TeamLoRA, integrates collaborative and competitive modules to improve performance. |
| Approach: | They propose to introduce task-specific LoRA as domain experts to improve learning efficiency . teamLoRA integrates collaborative and competition modules to improve model learning . |
| Outcome: | Experiments show that TeamLoRA improves performance in multi-task learning . teamLorea integrates collaborative and competitive modules to improve performance . |
LSRL: Process-Supervised GRPO on Latent Recurrent States Improves Mathematical Reasoning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Latent-recurrent language models solve tasks by iteratively refining hidden states rather than emitting chain-of-thought tokens. |
| Approach: | They propose a process-supervised variant of Guided Reward Policy Optimization that rewards latent steps at every latent step. |
| Outcome: | The proposed model improves absolute accuracy by +4.27 points on GSM-8K and +2.06 points on MathQA. |
LoRaDA: Low-Rank Direct Attention Adaptation for Efficient LLM Fine-tuning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Recent advances in parameter-efficient fine-tuning techniques allow for adjustments to only a minor fraction of the parameters of language models. |
| Approach: | They propose a low-rank direct attention adapted method for efficient LLM fine-tuning . they propose LMAM, which can bring negative attention to self-attention modules . |
| Outcome: | The proposed method outperforms the full fine-tuning method by 2.1% on GLUE benchmark. |
MoKA:Parameter Efficiency Fine-Tuning via Mixture of Kronecker Product Adaption (2025.coling-main)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) is one of the most popular PEFT methods . low-rank update mechanism of LoRA somewhat limits its ability to approximate full-parameter fine-tuning during training process. |
| Approach: | They propose a parameter-efficient fine-tuning framework that combines Kronecker product with the Mixture-of-Experts method to achieve parameter efficiency and better model performance. |
| Outcome: | The proposed framework outperforms existing methods on the GLUE benchmark and instruction tuning tasks for large language models. |
DisLoRA: Task-specific Low-Rank Adaptation via Orthogonal Basis from Singular Value Decomposition (2025.emnlp-main)
Copied to clipboard
| Challenge: | Extensive experiments on GLUE and Commonsense Reasoning benchmarks demonstrate that DisLoRA surpasses established PEFT methods, including LoRA, PiSSA, DoRA, LoRA-Dash, and SORSA. |
| Approach: | They propose a framework that leverages singular value decomposition to decompose pretrained weight matrices into orthogonal backbone and task-specific subspaces. |
| Outcome: | Extensive experiments on GLUE and Commonsense Reasoning benchmarks show that DisLoRA surpasses established PEFT methods, including LoRA, PiSSA, DoRA, LoRA-Dash, and SORSA. |
LoRA-Flow: Dynamic LoRA Fusion for Large Language Models in Generative Tasks (2024.acl-long)
Copied to clipboard
| Challenge: | LoRA-Flow uses lightweight modules to customize large language models for downstream tasks . previous work on LoRA combination relied on task-level weights for each involved LoRA . |
| Approach: | They propose a LoRA-Flow approach that uses dynamic weights to adjust the impact of different LoRAs. |
| Outcome: | The proposed method outperforms baselines with task-level weights on six generative tasks. |
Neeko: Leveraging Dynamic LoRA for Efficient Multi-Character Role-Playing Agent (2024.emnlp-main)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have revolutionized open-domain dialogue agents but face challenges in multi-character role-playing (MCRP) scenarios. |
| Approach: | They propose a framework for efficient multi-character role-playing that employs a dynamic low-rank adapter strategy and distinct LoRA blocks for each character. |
| Outcome: | Neeko employs a dynamic low-rank adapter (LoRA) strategy, enabling it to adapt seamlessly to diverse characters. |
Sensitivity-LoRA : Low-Load Sensitivity-Based Fine-Tuning for Large Language Models (2025.findings-emnlp)
Copied to clipboard
Hao Zhang, Bo Huang, Zhenjia Li, Xi Xiao, Hui Yi Leong, Zumeng Zhang, Xinwei Long, Tianyang Wang, Hao Xu
| Challenge: | Low-Rank Adaptation (LoRA) is a promising approach to adapting LLMs to specialized tasks . existing rank allocation techniques remain computationally inefficient and unstable . |
| Approach: | They propose a low-rank adapted model that approximates model weight updates using low-ranked decomposition. |
| Outcome: | The proposed method is limited by its uniform rank allocation to each incremental matrix . it leverages the second-order derivatives of the loss function to capture weight sensitivity . |
Flexora: Flexible Low-Rank Adaptation for Large Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) have revolutionized artificial intelligence, but performance on specific tasks is limited by knowledge boundaries. |
| Approach: | They propose a method that automatically selects the most critical layers for fine-tuning to optimize performance across diverse downstream tasks. |
| Outcome: | The proposed method outperforms baseline models and natural language tasks. |
Debiasing the Fine-Grained Classification Task in LLMs with Bias-Aware PEFT (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods to mitigate label biases such as retraining, post-hoc adjustment, and parameter-efficient fine-tuning fail to address prediction propensity and discriminative ability biase. |
| Approach: | They propose a bias-aware optimization framework that incorporates two distinct label balance constraints with a PEFT strategy targeting an intermediate layer to mitigate this issue. |
| Outcome: | The proposed approach outperforms or matches the performance of full-parameter fine-tuning and LoRA, achieving superior results with lower perplexity. |
Heterogeneous LoRA for Federated Fine-tuning of On-Device Foundation Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | federated fine-tuning of ODFMs is limited due to their limited size and system heterogeneity . emerging foundation models (FMs) have remarkable zero/few shot learning capabilities . |
| Approach: | They propose a federated fine-tuning method that leverages system and data heterogeneity at the edge. |
| Outcome: | a proposed method for federated fine-tuning improves performance on ODFMs . it allows heterogeneous LoRA ranks across clients for their individual system resources . |
Representation-Guided Parameter-Efficient LLM Unlearning (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to unlearning large language models often memorize sensitive or harmful information, but they struggle with the forget-retain trade-off due to the polysemantic nature of LLMs parameters. |
| Approach: | They propose a representation-guided low-rank unlearning approach that leverages the geometric properties of representation spaces to achieve robust and precise unlearning. |
| Outcome: | The proposed approach outperforms state-of-the-art models on TOFU and WMDP benchmarks while maintaining higher model utility. |
Mixture of LoRA Experts for Continual Information Extraction with LLMs (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods to continual information extraction are either task-specialized for a single task or suffer from catastrophic forgetting and insufficient knowledge transfer in continual IE. |
| Approach: | They propose a continual IE model that uses token-level mixture of LoRA experts with LLMs to extract emerging information across diverse IE tasks incessantly while avoiding forgetting. |
| Outcome: | The proposed model achieves state-of-the-art performance, effectively mitigating catastrophic forgetting and enhancing knowledge transfer in continual IE. |
Continual Gradient Low-Rank Projection Fine-Tuning for LLMs (2025.acl-long)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) offers efficiency but constrains the model’s ability to learn new tasks and transfer knowledge due to its low-rank nature and reliance on explicit parameter constraints. |
| Approach: | They propose a training strategy that synergistically combines full and low-rank parameters and jointly updating within a unified low-ranked gradient subspace. |
| Outcome: | Extensive experiments on continual learning benchmarks show that GORP improves performance compared to state-of-the-art approaches. |
CoLA: Collaborative Low-Rank Adaptation (2025.findings-acl)
Copied to clipboard
| Challenge: | The scaling law of Large Language Models (LLMs) reveals diminishing return on performance as model scale increases. |
| Approach: | They propose a more flexible LoRA architecture with an efficient initialization scheme . they propose combining three collaborative strategies to enhance performance . |
| Outcome: | The proposed model outperforms existing methods in low-sample scenarios. |
Advancing Parameter Efficiency in Fine-tuning via Representation Editing (2024.acl-long)
Copied to clipboard
Muling Wu, Wenhao Liu, Xiaohua Wang, Tianlong Li, Changze Lv, Zixuan Ling, Zhu JianHao, Cenyuan Zhang, Xiaoqing Zheng, Xuanjing Huang
| Challenge: | Parameter Efficient Fine-Tuning (PEFT) has gained significant attention for its ability to achieve competitive results while updating only a small subset of trainable parameters. |
| Approach: | They propose a new approach to fine-tuning neural models that scales and biases the representation produced at each layer. |
| Outcome: | The proposed approach reduces the number of trainable parameters by a factor of 25,700 compared to full parameter fine-tuning and by . 32 compared with LoRA. |
LoRA-PAR: A Flexible Dual-System LoRA Partitioning Approach to Efficient LLM Fine-Tuning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large-scale generative models like DeepSeek-R1 and OpenAI-O1 benefit substantially from chain-of-thought reasoning, yet pushing their performance typically requires vast data, large model sizes, and full-parameter fine-tuning. |
| Approach: | They propose a dual-system LoRA framework that partitions data and parameters by System 1 or System 2 demands and adopts a two-stage fine-tuning strategy to enhance knowledge and intuition. |
| Outcome: | The proposed framework partitions data and parameters by System 1 or System 2 demands, using fewer yet more focused parameters for each task. |
LLaVA Steering: Visual Instruction Tuning with 500x Fewer Parameters through Modality Linear Representation-Steering (2025.acl-long)
Copied to clipboard
| Challenge: | Multimodal Large Language Models (MLLMs) enhance visual tasks by integrating visual representations into large language models. |
| Approach: | They propose a method to re-balance modalities by steering visual representations . they propose LLaVA Steering, a platform that enables rapid customization of MLLMs a component-based architecture . |
| Outcome: | The proposed model re-balances the modalities of visual representations in large language models . the model requires 500 times fewer trainable parameters than LoRA while maintaining comparable performance . |
Steering Large Language Models for Machine Translation with Finetuning and In-Context Learning (2023.findings-emnlp)
Copied to clipboard
Duarte Alves, Nuno Guerreiro, João Alves, José Pombal, Ricardo Rei, José de Souza, Pierre Colombo, Andre Martins
| Challenge: | Large language models (LLMs) are a promising avenue for machine translation (MT) however, their effectiveness depends on the choice of few-shot examples and they often require extra post-processing due to overgeneration. |
| Approach: | They propose a method that incorporates few-shot examples during finetuning to improve performance on MT tasks. |
| Outcome: | The proposed method outperforms few-shot prompting while eliminating the need for in-context examples. |
LoRASC: Expressive and Generalizable Low-rank Adaptation for Large Models via Slow Cascaded Learning (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing low-rank adaptations have limited expressiveness, a tendency to overfit, and sensitivity to hyperparameter settings. |
| Approach: | They propose a technique to enhance LoRA’s expressiveness and generalization capabilities while preserving its training efficiency. |
| Outcome: | The proposed method outperforms baselines, mitigates overfitting, enhances model stability, and improves OOD robustness. |
Conan-Embedding-v2: Training an LLM from Scratch for Text Embeddings (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies use LoRA to fine-tune existing LLMs, but this is limited by the data and training gap between them and embedding models. |
| Approach: | They propose a new 1.4B-parameter LLM trained from scratch and fine-tuned as a text embedder that integrates embeddings across different languages. |
| Outcome: | The proposed model improves performance on the Massive Text Embedding Benchmark (MTEB) and Chinese MTEB (May 19, 2025). |
MLAS-LoRA: Language-Aware Parameters Detection and LoRA-Based Knowledge Transfer for Multilingual Machine Translation (2025.acl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) have demonstrated strong performance even with limited parallel data. |
| Approach: | They propose a multiple language-aware LoRA knowledge transfer framework that selectively adapts LLMs to MT by transferring knowledge from a large teacher to a small student model. |
| Outcome: | The proposed framework outperforms baseline models on multilingual language pairs by +1.7 BLEU on average. |
Multilingual Federated Low-Rank Adaptation for Collaborative Content Anomaly Detection across Multilingual Social Media Participants (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent developments in multilingual social media platforms (SNS) exacerbate new challenges in SNS content anomaly detection due to data islands and linguistic imbalance. |
| Approach: | They propose a multilingual Federated LoRA based on SVD-based language-specific disentanglement of LoRA blocks and a local orthogonal tuning strategy to detect content anomalies. |
| Outcome: | The proposed solution is superior in multilingual content anomaly detection while reducing multilingual knowledge conflicts and communication rounds. |
IAPT: Instance-Aware Prompt Tuning for Large Language Models (2024.acl-long)
Copied to clipboard
| Challenge: | Existing methods for prompt tuning require many soft tokens to guarantee performance . large language models still require a large amount of GPU memory and computations to fine-tune . |
| Approach: | They propose to use a parameter-efficient soft prompt generator to generate idiosyncratic soft prompts for each input instruction. |
| Outcome: | The proposed method outperforms the baselines with comparable tunable parameters and is more efficient than LoRA under the single-backbone multi-tenant setting. |
Weed Out, Then Harvest: Dual Low-Rank Adaptation is an Effective Noisy Label Detector for Noise-Robust Learning (2025.findings-acl)
Copied to clipboard
| Challenge: | Experimental results show that PEFT can fine-tune language models without relying on perfectly labeled datasets. |
| Approach: | They propose a framework that decouples sample selection from model training by introducing clean and noisy LoRA. |
| Outcome: | The proposed framework decouples sample selection from model training. |
PROPER: A Progressive Learning Framework for Personalized Large Language Models with Group-Level Adaptation (2025.acl-long)
Copied to clipboard
| Challenge: | Personalized large language models (LLMs) aim to tailor outputs to user preferences . however, user data is typically sparse, making it challenging to adapt LLMs to specific user patterns. |
| Approach: | They propose a progressive learning framework that groups users based on preferences and adapts LLMs in stages. |
| Outcome: | The proposed approach outperforms SOTA models across multiple tasks. |
CoMoL: Efficient Mixture of LoRA Experts via Dynamic Core Space Merging (2026.findings-acl)
Copied to clipboard
Jie Cao, Zhenxuan Fan, Zhuonan Wang, Tianwei Lin, Ziyuan Zhao, Rolan Yan, Wenqiao Zhang, Feifei Shao, Hongwei Wang, Jun Xiao, Siliang Tang
| Challenge: | Existing PEFT methods suffer from limited parameter efficiency and coarse-grained adaptation due to proliferation of LoRA experts and instance-level routing. |
| Approach: | They propose a new MoE-LoRA framework that incorporates expert diversity, parameter efficiency, and fine-grained adaptation. |
| Outcome: | The proposed framework outperforms existing methods on multiple tasks while maintaining parameter efficiency. |
Neuron-Level Differentiation of Memorization and Generalization in Large Language Models (2025.emnlp-main)
Copied to clipboard
Ko-Wei Huang, Yi-Fu Fu, Ching-Yu Tsai, Yu-Chieh Tu, Tzu-ling Cheng, Cheng-Yu Lin, Yi-Ting Yang, Heng-Yi Liu, Keng-Te Liao, Da-Cheng Juan, Shou-De Lin
| Challenge: | Existing models exhibit memorization and generalization behaviors in ways that are not easily interpretable or controllable. |
| Approach: | They propose to use a GPT-2 and LLaMA-3.2 model to identify distinct neuron subsets responsible for each behavior to steer the model toward memorization or generalization. |
| Outcome: | The proposed models show that inference-time interventions on these neurons can steer the model’s behavior toward memorization or generalization. |
LUNE: Efficient LLM Unlearning via LoRA Fine-Tuning with Negative Examples (2026.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models encode vast factual knowledge, yet their inability to selectively forget specific information hinders privacy protection, bias mitigation, and post-deployment correction. |
| Approach: | They propose a LoRA-based negative-only unlearning framework that updates only low-rank adapters while freezing the backbone. |
| Outcome: | The proposed framework reduces computational cost by about an order of magnitude compared to full fine-tuning and memory-editing methods. |
Adapting Where It Matters: Depth-Aware Adaptation for Efficient Multilingual Speech Recognition in Low-Resource Languages (2026.findings-acl)
Copied to clipboard
| Challenge: | Recent speech foundation models excel at multilingual automatic speech recognition (ASR) for high-resource languages, but their performance drops substantially on low-resourced languages due to the limited data availability. |
| Approach: | They propose a Depth-Aware Model Adaptation framework that allocates adaptation capacity according to each layer’s role. |
| Outcome: | The proposed framework matches or surpasses state-of-the-art accuracy with 80% fewer trainable parameters and achieves 29% error reduction under extreme data scarcity. |
SparseGrad: A Selective Method for Efficient Fine-tuning of MLP Layers (2024.emnlp-main)
Copied to clipboard
Viktoriia Chekalina, Anna Rudenko, Gleb Mezentsev, Aleksandr Mikhalev, Alexander Panchenko, Ivan Oseledets
| Challenge: | High-performance methods for parameter-efficient fine-tuning (PEFT) typically work with Attention blocks and overlook dense MLP blocks, which contain about half of the model parameters. |
| Approach: | They propose a selective PEFT method that performs well on MLP blocks by converting layer gradients into a sparse structure and reducing the number of updated parameters. |
| Outcome: | The proposed method outperforms LoRA and MeProp, robust state-of-the-art PEFT approaches. |
Boost, Disentangle, and Customize: A Robust System2-to-System1 Pipeline for Code Generation (2025.findings-acl)
Copied to clipboard
Kounianhua Du, Hanjing Wang, Jianxing Liu, Jizheng Chen, Xinyi Dai, Yasheng Wang, Ruiming Tang, Yong Yu, Jun Wang, Weinan Zhang
| Challenge: | Existing systems 2 methods for code generation are difficult to implement due to the complex hidden reasoning process and heterogeneous data distribution. |
| Approach: | They propose a framework that Boosts reasoning exploration via multi-agent collaboration and Disentangles heterogeneous data into specialized experts. |
| Outcome: | The proposed framework outperforms state-of-the-art methods on APPS and CodeContest benchmarks and achieves 73.8% accuracy on hard problems. |
TalkLoRA: Communication-Aware Mixture of Low-Rank Adaptation for Large Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Existing LoRA methods assume that experts operate independently, leading to unstable routing, expert dominance. |
| Approach: | They propose a communication-aware MoELoRA framework that relaxes this assumption by introducing expert-level communication prior to routing. |
| Outcome: | The proposed framework outperforms vanilla LoRA and MoELoRA on diverse language understanding tasks while maintaining expert dominance. |
TARE: Lightweight Token-Aware Representation Editing for Fine-tuning Transformer-like Models (2026.acl-long)
Copied to clipboard
| Challenge: | Existing PEFT methods can be costly and underfit token-level contexts. |
| Approach: | They propose a PEFT method that performs fine-grained, token-specific edits with a small additional inference overhead and minimal tuning. |
| Outcome: | The proposed method outperforms state-of-the-art methods in 8 tasks and GLUE with a minimal tuning overhead and inference overhead. |
RanLoRA: Residual-aware Nonlinear Low-Rank Adaptation (2026.findings-acl)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) relying on linear low-rank projections restricts adaptation to linear subspaces, limiting flexibility on complex downstream tasks. |
| Approach: | They propose a nonlinear low-rank Adaptation approach that leverages pretrained weights to decompose them into principal components that are kept frozen and residual components that can be used for task-specific adaptation. |
| Outcome: | The proposed approach outperforms vanilla LoRA and representative variants on commonsense reasoning, image classification, and mathematical reasoning tasks. |
Progressive Tuning: Towards Generic Sentiment Abilities for Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing models of sentiment understanding do not consider interrelated sentiment knowledge . et al., 2023; Zhao e.t., 20, 21; Shu e t. 2021) focus on individual sentiment subtasks . |
| Approach: | They propose an open-source large language model specific to the sentiment domain that explores hierarchical relationships between subtasks. |
| Outcome: | The proposed model performs well across all datasets in the progressive sentiment reasoning benchmark. |
Language Models Can Easily Learn to Reason from Demonstrations (2025.findings-emnlp)
Copied to clipboard
Dacheng Li, Shiyi Cao, Tyler Griggs, Shu Liu, Xiangxi Mo, Eric Tang, Sumanth Hegde, Kourosh Hakhamaneshi, Shishir G Patil, Matei Zaharia, Joseph E. Gonzalez, Ion Stoica
| Challenge: | Large reasoning models (LRMs) tackle complex problems by following long chain-of-thoughts (Long CoT) however, the training techniques and data requirements to elicit Long CoT remain poorly understood. |
| Approach: | They propose to use data-efficient supervised fine-tuning and parameter-efficient low-rank adaptation to elicit Long CoT reasoning. |
| Outcome: | The proposed model can learn Long CoT reasoning through data-efficient supervised fine-tuning and parameter-efficient low-rank adaptation. |
Low-Rank Interconnected Adaptation across Layers (2025.findings-acl)
Copied to clipboard
| Challenge: | Low-rank adaptation (LoRA) is a parameter-efficient fine-tuning (PEFT) method that learns weight updates W = AB for pretrained weights W through low-rank adapters A and B. |
| Approach: | They propose a low-rank interconnected adaptation across layers method that introduces an interconnected framework with locally shared A and globally shared B experts. |
| Outcome: | The proposed method improves expressiveness across domains and modalities and enables higher-rank W with equal or fewer parameters. |
ELBA-Bench: An Efficient Learning Backdoor Attacks Benchmark for Large Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Existing backdoor models are limited in coverage of attack, system integrity and backdoor alignment . ELBA-Bench provides over 1300 experiments encompassing 12 attack methods, 18 datasets, and 12 LLMs. |
| Approach: | They propose a framework that allows attackers to inject backdoor through parameter efficient fine-tuning or without fine-uning techniques. |
| Outcome: | ELBA-Bench provides over 1300 experiments encompassing 12 attack methods, 18 datasets, and 12 LLMs. |
FARSS: Fisher-Optimized Adaptive Low-Rank and Singular-Vector Selection for Knowledge-Preserving Fine-Tuning (2026.findings-acl)
Copied to clipboard
Renxing Chen, Ziwei Xiang, Peisong Wang, Hongjian Fang, Meng Li, Fanhu Zeng, Yanan Zhu, Peipei Yang, Xu-Yao Zhang, Jian Cheng
| Challenge: | Low-rank adaptation methods for large language models have limitations in preserving world knowledge and limiting updates to preserve world knowledge. |
| Approach: | They propose a Fisher-optimized adaptive low Rank and Singular-VectorSelection framework for knowledge-preserving fine-tuning that allows efficient and task-sensitive updates. |
| Outcome: | The proposed framework outperforms existing methods for knowledge-preserving fine-tuning. |
Advancing Vision-Language Models with Adapter Ensemble Strategies (2024.findings-emnlp)
Copied to clipboard
| Challenge: | CLIP revolutes vision-language pretraining by using contrastive learning on paired web data. |
| Approach: | They propose to combine a "adapter ensemble" with traditional machine learning techniques to augment large-scale pretrained vision-language models. |
| Outcome: | The proposed model outperforms baselines and derives improvement when the number of ensemble parameters increases. |
A Study of Parameter Efficient Fine-tuning by Learning to Efficiently Fine-Tune (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for parameter-efficient fine-tuning (PEFT) are limited due to the need for increased computational resources. |
| Approach: | They propose a method to learn PEFT parameters from data by projecting high dimensional parameters onto low dimensional parameter manifolds or identifying PEFT parametrically. |
| Outcome: | The proposed method can be used to identify PEFT parameters on GLUE tasks. |
RA-LoRA: Rank-Adaptive Parameter-Efficient Fine-Tuning for Accurate 2-bit Quantized Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) with their extensive parameters and high memory demands are challenging to fine-tune for specific applications with limited resources. |
| Approach: | They propose a method that dynamically adjusts the adapter’s rank using rank-subspace analysis, optimizing performance with fewer parameters. |
| Outcome: | The proposed method improves model accuracy with minimal parameter changes and demonstrates the importance of rank dynamics in optimizing quantized LLMs. |
Privacy-Preserving Reasoning with Knowledge-Distilled Parametric Retrieval Augmented Generation (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing RAG systems require uploading local documents to the cloud, resulting in inference latency and poor generalization on out-of-distribution (OOD) inputs. |
| Approach: | They propose a generalizable knowledge-distilled parametric RAG model aligned with standard RAG in document structure and parameter activation. |
| Outcome: | The proposed model outperforms baselines in accuracy and generalizes well on out-of-distribution (OOD) data. |
LoNAS: Elastic Low-Rank Adapters for Efficient Large Language Models (2024.lrec-main)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) reach hundreds of billions of parameters and require resources for training and inference stages. |
| Approach: | They propose a low-rank adapter to reduce the number of trainable parameters in a model and reduce memory requirements. |
| Outcome: | The proposed approach reduces memory and compute requirements while preserving performance. |
An Orthogonal High-Rank Adaptation for Large Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Low-rank adaptation (LoRA) efficiently adapts LLMs to downstream tasks by decomposing LLM’s weight update into trainable low-rank matrices for fine-tuning. |
| Approach: | They propose an orthogonal high-rank adaptation for parameter-efficient fine-tuning that decomposes LLMs’ pre-trained weight matrices into orthogonals via QR decomposition and splits them into two low-redundancy high-ranked components. |
| Outcome: | Empirical results show that OHoRA outperforms LoRA and its variants and generates task-tailored representation spaces with 0.0371% trainable parameters. |
Sci-LoRA: Mixture of Scientific LoRAs for Cross-Domain Lay Paraphrasing (2025.findings-acl)
Copied to clipboard
| Challenge: | Lay paraphrasing aims to make scientific information accessible to non-experts . existing studies focus on a single domain, such as biomedicine . |
| Approach: | a new lay paraphrasing model leverages a mixture of LoRAs fine-tuned on multiple scientific domains. |
| Outcome: | a new model outperforms state-of-the-art large language models in lay paraphrasing . the model can adjust the impact of different domains without explicit labels . |
Hebbian-Guided Bi-Directional Rank Adaptation for Parameter-Efficient Fine-Tuning (2026.findings-acl)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) is a widely used method to fine-tune large language models . but its fixed-rank design cannot capture the varying importance across different layers . |
| Approach: | They propose a framework that bi-directionally reallocates low-rank capacity using Hebbian-inspired importance estimation. |
| Outcome: | Experiments show that HeBiRA improves performance over baselines. |
MoA: Heterogeneous Mixture of Adapters for Parameter-Efficient Fine-Tuning of Large Language Models (2026.acl-long)
Copied to clipboard
Jie Cao, Tianwei Lin, Bo Yuan, Rolan Yan, Hongyang He, Wenqiao Zhang, Juncheng Li, Dongping Zhang, Siliang Tang, Yueting Zhuang
| Challenge: | Existing methods for parameter-efficient fine-tuning (PEFT) are limited by computational costs and performance degradation. |
| Approach: | They propose a method that integrates Low-Rank Adaptation and Mixture-of-Experts (MoE) they propose combining expert load imbalance and representation collapse to improve LLM performance . |
| Outcome: | The proposed method outperforms homogeneous MoE-LoRA architectures in performance and parameter efficiency. |
MolCA: Molecular Graph-Language Modeling with Cross-Modal Projector and Uni-Modal Adapter (2023.emnlp-main)
Copied to clipboard
| Challenge: | Language Models (LMs) have demonstrated impressive molecule understanding ability on 1D text-related tasks, but lack 2D graph perception, a critical ability of human professionals in comprehending molecules’ topological structures. |
| Approach: | They propose to combine a cross-modal projector and a uni-modal adapter to enable an LM to understand both text- and graph-based molecular contents via a Q-Former. |
| Outcome: | The proposed model outperforms the baselines on tasks such as molecule captioning, IUPAC name prediction, and molecule-text retrieval. |
From Style to Story: A Curriculum Learning Approach for Imitative Novel Generation (2026.findings-acl)
Copied to clipboard
| Challenge: | Novels create rich, immersive worlds with intricate plots and distinct styles, captivating readers through complex storytelling. |
| Approach: | They propose a novel generation system that imitates novel elements by predicting plot developments and writing concrete details using vivid, expressive language. |
| Outcome: | The novel imitative novel generation system is trained through a curriculum learning paradigm, progressing from low-level stylistic mastery to high-level narrative coherence. |
LLoCO: Learning Long Contexts Offline (2024.emnlp-main)
Copied to clipboard
Sijun Tan, Xiuyu Li, Shishir G Patil, Ziyang Wu, Tianjun Zhang, Kurt Keutzer, Joseph Gonzalez, Raluca Popa
| Challenge: | Large language models are still unable to handle long contexts due to the quadratic computational and memory overhead of the self-attention mechanism and the substantial KV cache sizes during generation. |
| Approach: | They propose a method to learn contexts offline through context compression and in-domain parameter-efficient finetuning with LoRA. |
| Outcome: | The proposed model outperforms in-context learning while using 30 fewer tokens during inference and significantly reduces the cost of long document question answering. |
Towards One-to-Many Visual Question Answering (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing Visual Question Answering systems are constrained to support domain-specific questions . a model trained on a single specific domain may not be competent for real-world application. |
| Approach: | They propose a task to enable a single model to answer as many different domains of questions as possible . they break the task down into the integration of three key abilities . |
| Outcome: | The proposed model can answer as many domains of questions as possible, the authors argue . the proposed model generalizes well to three extra zero-shot datasets, and the results are published. |
Adaptive LoRA Merge with Parameter Pruning for Low-Resource Generation (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for adapting LLMs to low-resource tasks keep LoRA parameters frozen and the low-level problem out of their scope. |
| Approach: | They propose a LoRA merge method that updates and prunes LoRA parameters through fine-tuning with minimal target task data. |
| Outcome: | The proposed method improves performance on a low-resource language generation task and improves on previous methods. |
Deputy: Accelerating Large Language Model Inference with Dynamic Low-Rank Substitution (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing dynamic schemes such as early-exit and layer-drop reduce FLOPs but break batch processing or introduce KV-cache inconsistency. |
| Approach: | They propose a dynamic low-rank substitution framework that employs a lightweight decision module at each layer to dynamically determine the execution branch for different tokens. |
| Outcome: | The proposed model reduces computation by approximately 40% compared to the original dense model while outperforming existing baseline methods. |
MiLoRA: Efficient Mixture of Low-Rank Adaptation for Large Language Models Fine-tuning (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Low-rank adaptation and its mixture-of-experts (MOE) methods are highly effective but introduce significant latency in multi-tenant settings due to the LoRA modules and MOE routers added to multiple linear modules. |
| Approach: | They propose a low-rank adaptation variant that considers each LoRA module as an expert and employs a prompt-aware routing mechanism. |
| Outcome: | Extensive analysis on commonsense reasoning tasks and math reasoning tasks show that MiLoRA outperforms strong PEFT baselines with comparable tunable parameter budgets. |
Mixture-of-LoRAs: An Efficient Multitask Tuning Method for Large Language Models (2024.lrec-main)
Copied to clipboard
| Challenge: | Instruction Tuning has the potential to stimulate or enhance specific capabilities of large language models. |
| Approach: | They propose a mixture-of-LoRAs architecture which is a parameter-efficient tuning method designed for multi-task learning with LLMs. |
| Outcome: | The proposed method can be iteratively adapted to a new domain, enabling quick domain-specific adaptation. |
Outlier-weighed Layerwise Sampling for LLM Fine-tuning (2025.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are a powerful tool for processing complex natural language processing tasks. |
| Approach: | They propose an approach to fine-tune LLMs with outliers and a gradient low-rank projection to increase the number of fine-sampled layers without a proportional increase in memory costs. |
| Outcome: | The proposed approach outperforms baseline approaches while being more memory efficient. |
S2ST-Omni: Hierarchical Language-Aware SpeechLLM Adaptation for Multilingual Speech-to-Speech Translation (2026.findings-acl)
Copied to clipboard
| Challenge: | S2ST-Omni integrates a speech-to-text frontend with a modular, plug-and-play text-tospeech backend. |
| Approach: | They propose a compositional S2ST framework that integrates a speech-to-text frontend with a modular, plug-and-play text-tospeech backend. |
| Outcome: | The proposed framework outperforms existing frameworks in translation and synthesis . it integrates a speech-to-text translation frontend with a plug-and-play text-tospeech backend . |
G-LoRA: Global-Local Decoupled Low-Rank Adaptation (2026.findings-acl)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) improves the fine-tuning efficiency and performance of large language models. |
| Approach: | They propose a low-rank adaptive approach that decomposes update matrix into global and local adapters and assigns them to local and global adapters. |
| Outcome: | The proposed method achieves up to 2.7% accuracy improvement over LoRA and its variants on commonsense reasoning, mathematical reasoning, and code generation. |
Sparsity May Be All You Need: Sparse Random Parameter Adaptation (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Parameter-Efficient Fine-Tuning (PEFT) methods aim at reducing computational and memory resources for fine-tuning large language models. |
| Approach: | They propose to train on a small number of parameters instead of all model parameters . they compare the method to LoRA and find it to be efficient . |
| Outcome: | The proposed method is competitive with LoRA when using a similar number of trainable parameters. |
Multilingual Sentence-T5: Scalable Sentence Encoders for Multilingual Applications (2024.lrec-main)
Copied to clipboard
| Challenge: | Prior work on multilingual sentence embedding has demonstrated that the efficient use of natural language inference data to build high-performance models can outperform conventional methods. |
| Approach: | They propose a multilingual sentence embedding model by extending an existing monolingual model by using the low-rank adaptation technique. |
| Outcome: | The proposed model outperforms the previous approach and shows that languages with fewer resources or those with less linguistic similarity to English benefit more from the parameter increase. |
EcoLoRA: Communication-Efficient Federated Fine-Tuning of Large Language Models (2025.emnlp-main)
Copied to clipboard
Han Liu, Ruoyao Wen, Srijith Nair, Jia Liu, Wenjing Lou, Chongjie Zhang, William Yeoh, Yevgeniy Vorobeychik, Ning Zhang
| Challenge: | Recurrent exchange of model updates in FL can result in prohibitively high communication costs, hindering the distributed learning process. |
| Approach: | They propose a federated fine-tuning framework that uses a round-robin segment sharing scheme to reduce network bandwidth and adaptive sparsification methods tailored to LoRA’s training dynamics. |
| Outcome: | The proposed framework reduces communication overhead without compromising performance on question-answering and value-alignment tasks. |
Layer-aware Dual-directional Modulation for Low-resource Machine Translation (2026.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have demonstrated excellent performance in Machine Translation (MT) however, a performance gap persists between high-resource and low-resourced languages due to imbalanced pre-training data. |
| Approach: | They propose a layer-wise metric to quantify the activation divergence between high- and low-resource languages. |
| Outcome: | The proposed model outperforms standard LoRA fine-tuning on Chinese-to-seven low-resource language translations. |
NLoPT: N-gram Enhanced Low-Rank Task Adaptive Pre-training for Efficient Language Model Adaption (2024.lrec-main)
Copied to clipboard
| Challenge: | Pre-trained Language Models (PLMs) have superior performance on downstream tasks . however, conventional TAPT adjusts all parameters of the PLMs, which distorts the learned generic knowledge embedded in the original PLM's weights. |
| Approach: | They propose a two-step n-gram enhanced low-rank task adaptive pre-training method to customize a PLM to the downstream task. |
| Outcome: | The proposed method improves performance on six datasets from four domains. |
NormAL LoRA: What is the perfect size? (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are crucial for enabling intelligent experiences across applications. |
| Approach: | They propose a low-rank adaptive localization method that uses rank-norm regularization to determine the optimal rank for each weight matrix. |
| Outcome: | NormAL LoRA reduces adapter parameters by 37% while preserving full fine-tuning performance. |
Instant Personalized Large Language Model Adaptation via Hypernetwork (2026.acl-long)
Copied to clipboard
Zhaoxuan Tan, Zixuan Zhang, Haoyang Wen, Zheng Li, Rongzhi Zhang, Pei Chen, Fengran Mo, Zheyuan Liu, Qingkai Zeng, Qingyu Yin, Meng Jiang
| Challenge: | Existing parameter-efficient fine-tuning methods require training a separate adapter for each user, making them computationally expensive and impractical for real-time updates. |
| Approach: | They propose a scalable framework that maps a user's profile directly to a full set of adapter parameters. |
| Outcome: | The proposed framework outperforms prompt-based personalization and OPPU while using substantially fewer computational resources at deployment. |
Flipping Knowledge Distillation: Leveraging Small Models’ Expertise to Enhance LLMs in Text Matching (2025.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have demonstrated remarkable capabilities in acquiring diverse knowledge, making them highly effective across a wide range of tasks. |
| Approach: | They propose a flipped knowledge distillation paradigm where LLM learns from SLM . they propose to reinterpret LLMs as encoder-decoder models using LoRA . |
| Outcome: | The proposed model has been deployed in an online application environment and validated on financial and healthcare benchmarks and real-world applications. |
GPT vs RETRO: Exploring the Intersection of Retrieval and Parameter-Efficient Fine-Tuning (2024.emnlp-main)
Copied to clipboard
| Challenge: | Pre-trained large language models can be used for specific tasks and unique information but lack the resources for extensive retraining. |
| Approach: | They propose to use PEFT methods to adapt large language models while minimizing compute requirements. |
| Outcome: | The proposed methods outperform GPT models in zero-shot settings but lag behind PEFT. |
zFLoRA: Zero-Latency Fused Low-Rank Adapters (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) are increasingly deployed with task-specific adapters catering to multiple downstream applications. |
| Approach: | They propose a low-latency fused low-rank adapter that introduces zero latency overhead on top of the base model. |
| Outcome: | The proposed adapter reduces the inference time of the model by 2.5x . the proposed adapters are tested on 18 different tasks on different platforms . |
TableLoRA: Low-rank Adaptation on Table Structure Understanding for Large Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Tabular data are crucial in many fields and their understanding by large language models (LLMs) under high parameter efficiency paradigm is important. |
| Approach: | They propose a module that uses 2D LoRA to encode low-rank information on cell positions to improve table serialization and representation of two-dimensional structured information within a one-dimensional sequence. |
| Outcome: | Experiments on four tabular-related datasets show that TableLoRA outperforms vanilla LoRA and surpasses table encoding methods tested in control. |
Safe-FedLLM: Delving into the Safety of Federated Large Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Existing work on federated learning for large language models (FL) addresses privacy and data-silo issues in the training of large language model training. |
| Approach: | They propose a probe-based defense framework for FedLLM that constructs defenses across three levels: Step-Level, Client-Level and Shadow-Level. |
| Outcome: | The proposed framework improves FedLLM's robustness against malicious clients while maintaining competitive performance on benign data. |
MobiLoRA: Accelerating LoRA-based LLM Inference on Mobile Devices via Context-aware KV Cache Optimization (2025.acl-long)
Copied to clipboard
| Challenge: | MobiLoRA focuses on optimizing the key-value (KV) caches due to the limited computing and memory resources of mobile devices. |
| Approach: | They propose to optimize the key-value caches due to limited computing resources . they propose similarity-aware delta encoding for semantic-level contexts . |
| Outcome: | The proposed model accelerates LoRA-based LLM inference by 57.6% on mobile devices. |
AlphaLoRA: Assigning LoRA Experts Based on Layer Training Quality (2024.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies combine LoRA with Mixture-of-Experts (MoE) to improve performance in Large Language Models. |
| Approach: | They propose a method to combine LoRA and Mixture-of-Experts (MoE) to improve performance in Large Language Models. |
| Outcome: | The proposed method reduces redundancy in LoRA experts within the MoE architecture, and improves training quality across layers. |
Riemannian Optimization for LoRA on the Stiefel Manifold (2025.findings-emnlp)
Copied to clipboard
| Challenge: | powerful, large language models (LLMs) present significant fine-tuning challenges due to their size. |
| Approach: | They propose to optimize LoRA’s B matrix by imposing explicit orthogonality constraints that achieve near-perfect orthogonal and full effective rank. |
| Outcome: | The proposed method outperforms AdamW and LoRA in terms of parameter efficiency and representational capacity. |
TASO: Task-Aligned Sparse Optimization for Parameter-Efficient Model Adaptation (2025.emnlp-main)
Copied to clipboard
Daiye Miao, Yufang Liu, Jie Wang, Changzhi Sun, Yunke Zhang, Demei Yan, Shaokang Dong, Qi Zhang, Yuanbin Wu
| Challenge: | Existing studies have shown that LoRA introduces substantial parameter redundancy, which not only increases the number of trainable parameters but also hinders the effectiveness of fine-tuning. |
| Approach: | They propose a method that leverages importance information from the pretrained model’s weights to mitigate LoRA redundancy. |
| Outcome: | The proposed method significantly reduces the number of trainable parameters required for task adaptation while providing a task-aligned perspective for LoRA redundancy reduction. |
GMFL: Efficient Global Masking for Federated LLM Fine-tuning (2026.acl-long)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) has emerged as a prominent solution to mitigate the communication and computation costs in federated fine-tuning of Large Language Models (LLMs). |
| Approach: | They propose a plug-and-play layer freezing mechanism to integrate with existing federated fine-tuning frameworks. |
| Outcome: | The proposed solution reduces communication overhead and lowers computational costs while preserving the performance of the underlying federated fine-tuning methods. |
Can Spectral-Clipping Enable Better Learning While Forgetting Less for Low-Rank Adaptation? (2026.acl-long)
Copied to clipboard
| Challenge: | In recent years, low-rank adaptation (LoRA) has emerged as a significant paradigm that freezes pre-trained weights and introduces small, learnable adapters instead of fine-tuning the full set of parameters. |
| Approach: | They propose a low-rank adaptation approach that injects parameterized singular components with spectral clipping into the pre-trained model. |
| Outcome: | The proposed method improves performance and retains pre-trained knowledge while preserving the weights of the model. |
Memory-Efficient Fine-Tuning of Transformers via Token Selection (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for fine-tuning require caching of intermediate activations to update weights during the backward pass. |
| Approach: | They develop a method to reduce memory usage in fine-tuning of transformers by backpropagating through just a subset of input tokens. |
| Outcome: | The proposed method reduces memory usage and memory footprint on large transformer models . it can be easily combined with existing methods like LoRA, reducing memory cost . |
SDC-LoRA: Singular-Subspace Drift Controlled LoRA to Mitigate Knowledge Forgetting (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to adapt LLMs to new tasks focus on limiting knowledge forgetting . et al., 2023b) suggest a solution to this problem by limiting update energy in the principal singular subspace of W0 . |
| Approach: | They propose a low-rank Adaptation (LoRA) that steers early updates away from principal directions and mitigates forgetting by constraining update energy in the principal singular subspace of W0. |
| Outcome: | The proposed model mitigates forgetting on MMLU, TruthfulQA, and HellaSwag while keeping minor-subspace updates unchanged. |
Localized Low-Rank Adaptation within Clustered Parameter Subspaces (2026.acl-long)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) for large language models has been successful in various domains. |
| Approach: | They propose to perform low-rank updates within clustered parameter subspaces . they group rows/columns of update matrix into locally coherent, uncorrelated subspace blocks . |
| Outcome: | Empirical results show that low-rank Adaptation (LoRA) is better than global adaptations in various domains. |
Efficient Ensemble for Fine-tuning Language Models on Multiple Datasets (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for fine-tuning language models are efficient when adapting to a single dataset. |
| Approach: | They propose to use an ensemble method for fine-tuning a language model to multiple datasets instead of a single adapter per task. |
| Outcome: | The proposed method improves performance on multiple datasets while preserving low-rank adaptation properties. |
LoRATK: LoRA Once, Backdoor Everywhere in the Share-and-Play Ecosystem (2025.findings-emnlp)
Copied to clipboard
Hongyi Liu, Shaochen Zhong, Xintong Sun, Minghao Tian, Mohsen Hariri, Zirui Liu, Ruixiang Tang, Zhimeng Jiang, Jiayi Yuan, Yu-Neng Chuang, Li Li, Soo-Hyun Choi, Rui Chen, Vipin Chaudhary, Xia Hu
| Challenge: | distributing LLMs without a proven track record like ‘meta-llama‘ or ‘qwen‘ rarely gains community traction. |
| Approach: | They propose a simple, efficient, yet specific recipe for a backdoor LoRA to be injected into task-enhancing LoRAs and examine the mechanisms of such infections. |
| Outcome: | The proposed model allows attackers to scale the distribution of compromised LoRAs with minimal effort by leveraging the rich pool of shared LoRA assets. |
Language Fusion for Parameter-Efficient Cross-lingual Transfer (2025.acl-long)
Copied to clipboard
| Challenge: | Limited availability of multilingual text corpora for pretraining results in poor performance on downstream tasks due to undertrained representation spaces for languages other than English. |
| Approach: | They propose a method that integrates source and target language representations within low-rank (LoRA) adapters using lightweight linear transformations to enhance representation quality and transfer performance for languages other than English. |
| Outcome: | The proposed method improves representation quality and performance for languages other than English while maintaining parameter efficiency. |
Unraveling LoRA Interference: Orthogonal Subspaces for Robust Model Merging (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for fine-tuning large language models fail due to performance degradation . existing methods fail for models fine- tuned with low-rank adaptation . |
| Approach: | They propose to constrain the LoRA subspace prior to fine-tuning to ensure that updates relevant to one task do not adversely shift outputs for others. |
| Outcome: | The proposed method can integrate with most existing merging algorithms, reducing unintended interference among tasks. |
Dissecting Clinical Reasoning in Natural Language Inference for Large Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Recent studies on large language models (LLMs) have demonstrated the impact of prompting strategies and fine-tuning techniques on their reasoning capabilities. |
| Approach: | They examine four classes of prompting strategies to elicit reasoning in large language models . they then construct demonstrations using a frontier model to distil multi-step reasoning capabilities into smaller models based on Low-Rank Adaptation (LoRA). |
| Outcome: | The proposed model improves in 75% of the models on MedNLI and TREC Clinical Trials. |
Context-Conditioned Masked LoRA: Dynamic Rank Routing for Compute-Efficient Parameter-Efficient Fine-Tuning (2026.findings-acl)
Copied to clipboard
| Challenge: | Large pretrained language models (LMs) are commonly adapted via fine-tuning, but full updates are costly at scale. |
| Approach: | They propose a lightweight router that activates an input-dependent subset of LoRA rank directions and turns it into dynamic rank routing. |
| Outcome: | The proposed method improves accuracy–efficiency Pareto frontier versus static-rank LoRA and adaptive-rank baselines, while preserving memory and reducing overhead. |
LaCo: Layer-wise Compensation for Pruned Large Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for predicting performance degradations of Large Language Models (LLMs) neglect the structural distortions caused by sparsity. |
| Approach: | They propose a framework that reorients the recovery paradigm from global adaptation to hierarchical representation alignment by sequentially optimizing each layer to reconstruct the model's hidden states. |
| Outcome: | The proposed framework surpasses parameter-efficient baselines in perplexity reduction and zero-shot reasoning. |
TLoRA: Task-aware Low Rank Adaptation of Large Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Existing low-rank Adaptation (LoRA) methods address only one factor, often at the cost of increased training complexity or reduced practical efficiency. |
| Approach: | They propose a low-rank Adaptation framework that optimizes initialization and resource allocation at the outset of training. |
| Outcome: | The proposed framework performs excellently across various tasks while reducing the number of trainable parameters. |
GeLoRA: Geometric Adaptive Ranks For Efficient LoRA Fine-tuning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing adaptive LoRA methods lack a theoretical foundation to guide this trade-off optimally. |
| Approach: | They propose a principled approach that estimates the intrinsic dimensionality of hidden data representations to adaptively select LoRA ranks. |
| Outcome: | Experiments show that GeLoRA outperforms adaptive LoRA methods by up to +1.0% . |
V-RoLoRA: RLVR-Driven MoE Routing for Steerable Pluralistic Alignment (2026.findings-acl)
Copied to clipboard
| Challenge: | Current methods for steering large language models rely on prompt engineering or reasoning-time guidance. |
| Approach: | They propose a value-controllable pluralistic alignment framework enhanced with conditioned gating that dynamically directs the flow among multiple experts based on an input value or moral vector. |
| Outcome: | The proposed method outperforms prompt-based steering and multi-task PEFT benchmarks on two 8-billion-parameter backbones. |
Hierarchical-Task-Aware Multi-modal Mixture of Incremental LoRA Experts for Embodied Continual Learning (2025.acl-long)
Copied to clipboard
| Challenge: | Existing continual learning setups for embodied intelligence focus on executing low-level actions, neglecting the ability to learn high-level planning and multi-level knowledge. |
| Approach: | They propose a Hierarchical Embodied Continual Learning Setups (HEC) that divides the agent’s continual learning process into two layers: high-level instructions and low-level actions. |
| Outcome: | The proposed method reduces the forgetting of old tasks compared to other methods, while orthogonally training the remaining parts. |
Reusable Experiences: Latent Routing and Modular Composition in LLMs (2026.acl-long)
Copied to clipboard
| Challenge: | Existing approaches represent accumulated experience as explicit textual artifacts in prompts or implicitly within model weights via fine-tuning. Existing methods are limited by context windows and cannot internalize knowledge. |
| Approach: | They propose a framework that treats latent experiences as fundamental units for LLM specialization. |
| Outcome: | Experiments on multi-task NLP benchmarks show that this approach outperforms standard fine-tuning, yielding improved generalization through flexible skill reuse. |
SAMoRA: Semantic-Aware Mixture of LoRA Experts for Task-Adaptive Learning (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for multitask learning fail to match input semantics with expert capabilities, leading to weak expert specialization. |
| Approach: | They propose a parameter-efficient mixture-of-experts framework for task-adaptive learning that aligns textual semantics with the most suitable experts for precise routing. |
| Outcome: | The proposed framework outperforms the state-of-the-art methods and holds excellent task generalization capabilities. |
From Outcome to Process: Optimizing MoE Load Balancing with MCTS (2026.findings-acl)
Copied to clipboard
Wenjun Ke, Hengyuan Xu, Ziyu Shang, Yao He, Jiahao Wang, Zijie Xu, Peng Wang, Yuhang Lou, Jiajun Liu
| Challenge: | Existing balancing strategies focus on constraining the final distribution of expert usage, but overlook the routing decisions made at each layer. |
| Approach: | They propose a three-stage framework that leverages process-level rewards to guide balanced expert routing. |
| Outcome: | Extensive experiments show that LayerMoE improves the performance of state-of-the-art LoRA-MoA baselines, yielding an average accuracy gain of 1.39%. |
PASs-MoE: Mitigating Misaligned Co-drift among Router and Experts via Pathway Activation Subspaces for Continual Learning (2026.acl-long)
Copied to clipboard
| Challenge: | Existing LoRA-based Mixture-of-Experts (MoE) methods often jointly update the router and experts in an indiscriminate way, causing the router’s preferences to co-drift with experts’ adaptation pathways and exacerbate forgetting. |
| Approach: | They propose a LoRA-induced subspace that reflects which low-rank pathway directions an input activates in each expert, providing a capability-aligned coordinate system for routing and preservation. |
| Outcome: | The proposed method outperforms conventional continual learning baselines and MoE–LoRA variants in accuracy and resistance to forgetting, without increasing model parameters. |
SCRIBE: Structured Chain Reasoning for Interactive Behaviour Explanations using Tool Calling (2025.emnlp-main)
Copied to clipboard
| Challenge: | Language models can be used to provide personalized feedback in educational settings, but they face privacy concerns, limited computational resources, and the need for pedagogically valid responses. |
| Approach: | They propose a framework for multi-hop, tool-augmented reasoning to generate valid responses to student questions about feedback reports. |
| Outcome: | The proposed framework can generate valid responses to student questions about feedback reports using domain-specific tools and self-reflective inference pipelines. |
Two-Stage Parameter Alignment for Multi-LoRA Merging in Large Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Current general model merging methods are prone to parameter interference problems . a novel two-stage parameter alignment framework is proposed to address this problem . |
| Approach: | They propose a two-stage parameter alignment framework that integrates low-rank LoRAs . they propose to reduce the computational complexity of existing methods by preserving fine-grained functions . |
| Outcome: | The proposed framework exhibits greater robustness than other methods in high-rank and high-interference scenarios while preserving fine-grained functions. |
Refining Salience-Aware Sparse Fine-Tuning Strategies for Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for fine-tuning large language models require expensive training on consumer-grade hardwares. |
| Approach: | They propose a sparsity-based approach that introduces trainable sparse adaptations to the weight matrices in the model and offers greater flexibility in selecting fine-tuned parameters. |
| Outcome: | The proposed method outperforms other methods for a simple yet effective baseline for nLP tasks while sacrificing performance. |
Improving Rule-based Reasoning in LLMs using Neurosymbolic Representations (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) face challenges in reliably solving reasoning tasks, especially when solving tasks that require strict rule following. |
| Approach: | They propose a method that encodes hidden states into neurosymbolic vectors and decodes them into a neurosample vector space to enable problem-solving within a neural space. |
| Outcome: | The proposed method shows an average of 88.6% lower cross-entropy loss and 15.4 times more problems correctly solved on a suite of mathematical reasoning tasks compared to chain-of-thought prompting and supervised fine-tuning (LoRA). |
More Thinking, Less Talking: Internalizing Deliberative Safety into LLM Parameters (2026.acl-long)
Copied to clipboard
| Challenge: | Existing safety alignment methods leave Large Language Models vulnerable to sophisticated jailbreak attacks. |
| Approach: | They propose a safety reasoning internalization framework that internalizes safety reasoning into an implicit computational pathway using Low-Rank Adaptation (LoRA). |
| Outcome: | The proposed framework achieves a 43% lower Attack Success Rate (ASR) against distinct jailbreak attacks compared to strong baselines. |
Grouped Adaptive Weight Sharing (GAWS): An Inference-Efficient Adaptation Method for Large Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) is a new approach to fine-tuning large language models . adapters are lightweight, task specific modules that can be used for adapters in latency-sensitive settings. |
| Approach: | They propose a low-rank adapter with a weight sharing mechanism that reduces latency by 40% . they analyze LoRA adapters on GPUs and identify segmented function calls as the primary source of latency. |
| Outcome: | The proposed adapter reduces latency to about 40% of the gap between the unmerged LoRA and the base model while maintaining parameter efficiency and comparable accuracy. |
Astra: Activation-Space Tail-Eigenvector Low-Rank Adaptation of Large Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for fine-tuning pre-trained models are limited due to suboptimal activation subspaces. |
| Approach: | They propose a method that leverages tail eigenvectors of model output activations to construct low-rank adapters. |
| Outcome: | The proposed method outperforms existing methods across 16 benchmarks and surpasses full fine-tuning in certain scenarios. |
LoRACoE: Improving Large Language Model via Composition-based LoRA Expert (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies show that the Mixture of Experts architecture improves performance of large language models. |
| Approach: | They propose a method to build static experts using LoRA parameters . they propose to use rank-level parameters to build experts based on rank-based parameters based in LoRA module. |
| Outcome: | The proposed method improves task performance across a broader range of tasks. |
Can Small Vision–Language Models Perform Sign Language Translation? (2026.findings-acl)
Copied to clipboard
| Challenge: | Vision-Language Models (VLMs) have shown strong generalization across multimodal tasks, but their capacity to handle sign language translation (SLT) remains unclear. |
| Approach: | They propose entity- and semantics-aware metrics tailored for SLT to evaluate their performance. |
| Outcome: | The proposed metrics highlight the limitations of general-purpose VLMs to SLT, unlike their applicability in other tasks. |
HyperAdaLoRA: Accelerating LoRA Rank Allocation During Training via Hypernetworks without Sacrificing Performance (2026.findings-acl)
Copied to clipboard
Hao Zhang, Zhenjia Li, Yifan Gao, Xi Xiao, Heng Zhang, Shuyang Zhang, null Xiaoxincc, Bo Huang, Yuhang Wu, Tianyang Wang, Hao Xu
| Challenge: | Low-Rank Adaptation (LoRA) assumes a uniform rank r for each incremental matrix, not accounting for the varying significance of weight matrices across modules and layers. |
| Approach: | They propose a framework that allows for faster convergence of low-rank adaptive models . they use a hypernetwork to prune the outputs of the hypernetworks to generate parameters . |
| Outcome: | The proposed framework accelerates convergence of AdaLoRA by leveraging a hypernetwork. |
DPLoRA: A Dual-Pruning Framework based on ILP Optimization and Progressive Pruning for Parameter-Efficient LoRA Fine-Tuning (2026.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) require computational resources for fine-tuning. |
| Approach: | They propose a framework that optimizes rank allocation via two stages . they propose an initial pruning stage and a progressive pruning stage . |
| Outcome: | The proposed framework outperforms existing PEFT baselines on GLUE and instruction-following tasks while reducing training time and trainable parameters by over 80%. |
Evolutionary Negative Module Pruning for Better LoRA Merging (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for integrating multiple low-rank Adaptation experts into a single backbone are limited by negative modules. |
| Approach: | They propose a plug-and-play LoRA pruning method to locate and exclude negative modules prior to merging. |
| Outcome: | The proposed method boosts the performance of existing merging algorithms across languages and vision domains. |
Why LoRA Fails to Forget: Regularized Low-Rank Adaptation Against Backdoors in Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) is widely used for parameter-efficient fine-tuning of large language models, but is ineffective at removing backdoor behaviors from poisoned pretrained models when fine-timing on clean datasets. |
| Approach: | They propose a low-rank Adaptation method which increases spectral strength and corrects alignment through clean-strengthened regularization and trigger-insensitive constraints. |
| Outcome: | The proposed method significantly reduces attack success rates while maintaining clean accuracy. |
Trait Activation in Silicon: A Situation-Aware Framework for Psychologically Grounded Role-Playing (2026.acl-long)
Copied to clipboard
| Challenge: | Role-playing agents lack a deep understanding of complex human psychological mechanisms. |
| Approach: | They propose a situation-aware framework that decouples personality traits into bidirectional LoRA adapters. |
| Outcome: | Empirical results show that PD-LLM achieves superior performance in both static fidelity and dynamic adaptability. |
LoRA on the Go: Instance-level Dynamic LoRA Selection and Merging (2026.acl-long)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) is a parameter-efficient approach for fine-tuning large language models. |
| Approach: | They propose a low-rank Adaptation framework that automatically selects and merges LoRA adapters at the instance level without additional training. |
| Outcome: | The proposed framework outperforms training-based baselines on some tasks upto a margin of 3.6% while remaining competitive on other tasks and maintaining inference throughput. |
GROLE: Instance-Level Group Relative Optimization for LoRA Experts in Incremental Learning (2026.findings-acl)
Copied to clipboard
Yongyi Liao, Wencan Lai, Jun Fang, Jinjin Guo, Xiaohui Zhang, Zhiyuan Liu, Chao Liu, Pengzhang Liu, Qixia Jiang
| Challenge: | Large language models demonstrate remarkable zero-shot generalization, but adapting to downstream tasks requires continual fine-tuning. |
| Approach: | They propose a method that incrementally constructs a pool of frozen, task-specific LoRA experts. |
| Outcome: | The proposed approach outperforms state-of-the-art methods in task-free and blurred-boundary settings. |
SecureGate: Learning When to Reveal PII Safely via Token-Gated Dual-Adapters for Federated LLMs (2026.acl-long)
Copied to clipboard
| Challenge: | Existing privacy defenses reduce leakage of PII due to LLM memorization, but often degrade downstream performance. |
| Approach: | They propose a privacy-aware federated fine-tuning framework for large language models that provides fine-grained privacy control without sacrificing utility. |
| Outcome: | The proposed framework reduces PII leakage while providing fine-grained privacy control without sacrificing utility. |
BanHADEX: Towards Explainable HAte Speech Detection in Bangla Using Human Annotated EXplanation (2026.acl-long)
Copied to clipboard
Faisal Hossain Raquib, Akm Moshiur Rahman Mazumder, Md Fahim, Md Tahmid Hasan Fuad, Md Farhan Ishmam, Faria Sultana, M Ashraful Amin, Amin Ahsan Ali, Akmmahbubur Rahman
| Challenge: | Existing studies in Bangla focus on hate classification while overlooking interpretability. |
| Approach: | They propose to create a dataset with human-annotated labels for banla that contains 19,203 YouTube comments spanning April 2024–June 2025. |
| Outcome: | The proposed dataset outperforms existing datasets on open and closed-source LLMs on interpretability and better understanding of hate speech in linguistically rich yet under-resourced languages. |
Multiplication in Multimodal LLMs: Computation with Text, Image, and Audio Inputs (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing benchmarks lack systematically paired instances across modalities, making it difficult to compare genuine arithmetic limits . a model that computes 4736 may fail on a nearby instance like 8967, despite a well-tuned internal router. |
| Approach: | They propose a controlled multimodal multiplication benchmark that factorially varies digit length, digit sparsity, representation, and modality with paired instances from a reproducible generator. |
| Outcome: | The proposed model can perceive numerical content across modalities but fails to perform exact multi-digit multiplication when presented as numerals, number words, images, or in audio form. |
MTA:A Merge-then-Adapt Framework for Personalized Large Language Models (2026.acl-long)
Copied to clipboard
Xiaopeng Li, Yuanjin Zheng, Wanyu Wang, Wenlin Zhang, Pengyue Jia, Yingyi Zhang, Haiying He, Mengyang Ma, Yiqi Wang, Maolin Wang, Xuetao Wei, Xiangyu Zhao
| Challenge: | Personalized Large Language Models (PLLMs) aim to align outputs with individual user preferences . current methods of fine-tuning a separate module for each user are unscalable . |
| Approach: | They propose a Merge-then-Adapt framework for Personalized Large Language Models . they construct a shared Meta-LoRA bank and propose an Adaptive LoRA Fusion stage . |
| Outcome: | The proposed framework outperforms existing SOTA methods on the LaMP benchmark. |
SHARP: Self-adaptive Harmful Category-aware Prompt Generation for Black-box Jailbreaking (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for jailbreak ignore the semantic differences between categories of harmful questions, leading to inconsistent success rates and reduced overall attack effectiveness. |
| Approach: | They propose a category-aware jailbreak framework that incorporates the semantic category of harmful questions into prompt generation. |
| Outcome: | The proposed framework improves attack success rates and category alignment and achieves better cross-category robustness compared to the state-of-the-art (SOTA) baselines. |
Red-Bandit: Test-Time Adaptation for LLM Red-Teaming via Bandit-Guided LoRA Experts (2026.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to audit Large Language Models (LLMs) lack mechanisms to efficiently adapt to model-specific vulnerabilities at inference. |
| Approach: | They propose a red-teaming framework that adapts online to identify and exploit model failure modes under distinct attack styles. |
| Outcome: | The proposed framework outperforms state-of-the-art methods on AdvBench and HarmBench, while generating more human-readable adversarial prompts (lower perplexity). |