Papers with expert models
Accelerating LLM Fine-Tuning via Embedding Knowledge Transfer (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing studies on parameter-efficient fine-tuning (PEFT) have produced many state-of-the-art results by adapting LLMs to new tasks, but it requires substantial training data and time to enhance model performance. |
| Approach: | They propose a parameter-efficient fine-tuning framework which efficiently transfers knowledge from a small expert model to a target large model via embedding layers. |
| Outcome: | The proposed framework accelerates domain-specific fine-tuning, improves model performance and remains robust across diverse model families and PEFT methods. |
Audio Query Handling System with Integrated Expert Models and Contextual Understanding (2025.emnlp-industry)
Copied to clipboard
| Challenge: | Existing chatbots are limited to specific audio tasks, but the domain of audio content related queries remains underexplored. |
| Approach: | They propose to use an intent classifier to route queries to audio-related experts using a diverse audio query dataset. |
| Outcome: | The proposed system outperforms state-of-the-art LLMs on custom audio tasks and MMAU sound set benchmarks. |
Combining Domain and Alignment Vectors Provides Better Knowledge-Safety Trade-offs in LLMs (2025.acl-short)
Copied to clipboard
| Challenge: | Large language models (LLMs) excel in specific technical fields, but are not explicitly trained to be safe. |
| Approach: | They propose a model merging-based alignment method that allows for safer domain-specific models that preserve their utility. |
| Outcome: | The proposed method improves safety alignment on LLMs with minimal degradation on domain-specific benchmarks. |
MoFE: Mixture of Frozen Experts Architecture (2025.naacl-industry)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are characterized by their immense size, often consisting of at least one billion parameters. |
| Approach: | They propose a mixture of Frozen Experts architecture that integrates PEFT and MoE to enhance both training efficiency and model scalability. |
| Outcome: | The proposed architecture outperforms other methods while achieving the highest efficiency. |
Two Heads are Better than One: Nested PoE for Robust Defense Against Multi-Backdoors (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing defense mechanisms assume that only one type of trigger is adopted by the attacker, while defending against multiple simultaneous and independent trigger types necessitates general defense frameworks. |
| Approach: | They propose a framework that uses a mixture of experts as a trigger-only ensemble to defend against multiple trigger types. |
| Outcome: | The proposed framework defends against multiple trigger types in a single ensemble and in combination of models. |
Moral Stories: Situated Reasoning about Norms, Intents, Actions, and their Consequences (2021.emnlp-main)
Copied to clipboard
| Challenge: | aaron carroll: in social settings, human behavior is governed by unspoken rules of conduct rooted in societal norms . carroll and colleagues examine whether language generation models can serve as behavioral priors if they are not . they say we examine whether they can generate descriptions of actions that accomplish predefined goals . |
| Approach: | They propose to combine multiple expert models to improve quality of generated actions, consequences, and norms. |
| Outcome: | The proposed models significantly improve the quality of generated actions, consequences, and norms compared to baselines. |
MergeME: Model Merging Techniques for Homogeneous and Heterogeneous MoEs (2025.naacl-long)
Copied to clipboard
Yuhang Zhou, Giannis Karamanolakis, Victor Soto, Anna Rumshisky, Mayank Kulkarni, Furong Huang, Wei Ai, Jianhua Lu
| Challenge: | State-of-the-art methods for merging expert models with different architectures do not address parameter interference and require extensive fine-tuning to restore performance. |
| Approach: | They propose a method for merging experts with different architectures into a unified Mixture-of-Experts model with a goal of enhancing performance in each domain while retaining effectiveness on general tasks. |
| Outcome: | Experiments across multiple domains show that the proposed methods reduce fine-tuning costs and improve performance over state-of-the-art methods. |
Think Outside the Policy: In-Context Steered Policy Optimization (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing Reinforcement Learning from Verifiable Rewards (RLVR) methods exhibit limited exploration due to reliance on on-policy rollouts which are limited to the current policy’s distribution, resulting in narrow trajectory diversity. |
| Approach: | They propose a framework that leverages the in-context learning capability of Large Reasoning Models to provide expert guidance using existing datasets. |
| Outcome: | The proposed framework improves RLVR performance and training stability on mathematical reasoning benchmarks. |
RADAR: Risk-Aware Distilled Adaptive Routing for Efficient Short-Form Video Platform Ecosystem Governance (2026.acl-industry)
Copied to clipboard
| Challenge: | Existing solutions to address inefficiency in large-scale integrity enforcement on short-form video platforms require multiple specialized vertical modules . |
| Approach: | They propose a lightweight risk-aware routing framework that selectively releases low-risk content while dispatching high-risk instances to appropriate vertical modules. |
| Outcome: | The proposed framework selectively releases low-risk content while dispatching high-risk instances to appropriate vertical modules. |
Towards Adaptive Mechanism Activation in Language Agent (2025.coling-main)
Copied to clipboard
| Challenge: | Existing Language Agents rely on a fixed mechanism or a set of mechanisms activated in a predefined order, limiting their adaptation to varied potential task solution structures. |
| Approach: | They propose to use language agents to learn to activate different mechanisms without relying on expert models to optimize their adaptation to different task solutions. |
| Outcome: | The proposed approach improves agent performance by enabling it to activate the appropriate mechanisms according to the potential characteristics of the task. |
Bridging the Gap between Expert and Language Models: Concept-guided Chess Commentary Generation and Evaluation (2025.naacl-long)
Copied to clipboard
| Challenge: | Experimental results show that expert models generate accurate, informative and fluent commentary, but are prone to hallucinations due to their limited decision-making capabilities. |
| Approach: | They propose a concept-guided chess commentary generation and a GPT-based Chess Commentary Evaluation to bridge this gap between expert models and large language models. |
| Outcome: | The proposed model is accurate, informative, and fluent. |
MMoE: Enhancing Multimodal Models with Mixtures of Multimodal Interaction Experts (2024.emnlp-main)
Copied to clipboard
| Challenge: | Multimodal models focus on the correspondence between images and text, but this only covers a subset of real-world interactions. |
| Approach: | They propose an approach to enhance multimodal models by training separate expert models for each type of interaction, such as redundancy present in both modalities, uniqueness in one modality, or synergy that emerges when both . modality is used to capture overlaps in semantic content between images and text, making a strong multi-view redundancies assumption. |
| Outcome: | The proposed approach improves on a sarcasm detection and humor detection task. |
PANDA: Preference Adaptation for Enhancing Domain-Specific Abilities of LLMs (2024.findings-acl)
Copied to clipboard
| Challenge: | Large language models have demonstrated considerable capabilities across various tasks . however, they often fall short of the performance achieved by domain-specific state-of-the-art models . |
| Approach: | They propose a tuning-free method to augment domain-specific abilities of Large language models . they leverage insights from the response preference of expert models to augment LLMs . |
| Outcome: | The proposed method outperforms the expert model on 4 ScienceWorld tasks. |
Are You Looking? Grounding to Multiple Modalities in Vision-and-Language Navigation (P19-1)
Copied to clipboard
| Challenge: | Existing models that ground language into visual appearance and route structure are outperforming their visual counterparts in unseen new environments. |
| Approach: | They propose to decompose the grounding procedure into a set of expert models with access to different modalities and ensemble them at prediction time. |
| Outcome: | The proposed model outperforms models with only route structure and visual features on the benchmark Room-to-Room dataset. |
SIFT-50M: A Large-Scale Multilingual Dataset for Speech Instruction Fine-Tuning (2025.acl-long)
Copied to clipboard
Prabhat Pandey, Rupak Vignesh Swaminathan, K V Vijay Girish, Arunasish Sen, Jian. Xie, Grant Strimel, Andreas Schwarz
| Challenge: | Recent years have witnessed significant advancements in integrating speech and audio capabilities into large language models. |
| Approach: | They propose a 50M-example dataset for instruction fine-tuning and pre-training of speech-text large language models (LLMs) the dataset spans five languages and enables a diverse range of speech understanding and controllable speech generation instructions. |
| Outcome: | The proposed dataset outperforms existing speech-text LLMs on instruction-following benchmarks while achieving competitive performance on foundational speech tasks. |
Debating for Better Reasoning in Vision-Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) gain expertise across diverse domains and modalities, a new study shows . scalable oversight becomes challenging when their capabilities surpass human evaluators. |
| Approach: | a new study extends the debate paradigm to a multimodal setting . it explores the potential for blind models to supervise and enhance the performance of sighted ones. |
| Outcome: | The proposed framework outperforms individual LLMs on multimodal tasks . it allows blind models to supervise and enhance the performance of sighted models . |
CAML: A Conflict-Aware Molecular Language Model Merging Framework for Multi-Constraint Molecular Generation (2026.acl-long)
Copied to clipboard
Xuanbai Ren, Luoda Tan, Pei Liu, Tengfei Ma, Xiangzheng Fu, Longyue Wang, Yiping Liu, Xiangxiang Zeng
| Challenge: | Existing paradigms struggle with this challenge due to catastrophic forgetting or gradient conflicts. |
| Approach: | They propose a conflict-aware molecular language model merging framework that generates multiple constraints moleculaire as a cooperative game among property-specific fine-tune models. |
| Outcome: | The proposed framework generates multiple constraints molecular as a cooperative game among property-specific fine-tune models (expert models) it minimizes conflicts among properties by exploring the optimal combination of the importance of the task parameter and relative fusion weights of each expert (fusion coefficient). |
A Modular Approach for Clinical SLMs Driven by Synthetic Data with Pre-Instruction Tuning, Model Merging, and Clinical-Tasks Alignment (2025.acl-long)
Copied to clipboard
Jean-Philippe Corbeil, Amin Dada, Jean-Michel Attendu, Asma Ben Abacha, Alessandro Sordoni, Lucas Caccia, Francois Beaulieu, Thomas Lin, Jens Kleesiek, Paul Vozila
| Challenge: | Large language models such as GPT-4 have limited their deployment in clinical settings . a novel framework for adapting SLMs into high-performing clinical models is needed . |
| Approach: | They propose a framework for adapting large language models into high-performing clinical models . they pre-instruct experts on relevant medical and clinical corpora and model merging . |
| Outcome: | The proposed framework outperforms the existing model on the CLUE+ benchmark on medical entities and radiology reports. |
AEGIS: A Holistic Benchmark for Evaluating Forensic Analysis of AI-Generated Academic Images (2026.acl-long)
Copied to clipboard
Bo Zhang, Tzu-Yen Ma, Zichen Tang, Junpeng Ding, Zirui Wang, Yizhuo Zhao, Peilin Gao, Zijie Xi, Zixin Ding, Haiyang Sun, Haocheng Gao, Yuan Liu, Liangjia Wang, Yiling Huang, Yujie Wang, Yuyue Zhang, Ronghui Xi, Yuanze Li, Jiacheng Liu, Zhongjun Yang, Haihong E
| Challenge: | AEGIS examines whether current models can effectively audit AI-generated images in academic papers. |
| Approach: | They propose a holistic benchmark for forensic analysis of AI-Generated academic ImageS that reveals limitations in academic image forensics. |
| Outcome: | AEGIS compared with existing benchmarks on seven academic categories and features key advances in forensic analysis. |
Dolphin: Document Image Parsing via Heterogeneous Anchor Prompting (2025.findings-acl)
Copied to clipboard
Hao Feng, Shu Wei, Xiang Fei, Wei Shi, Yingdong Han, Lei Liao, Jinghui Lu, Binghong Wu, Qi Liu, Chunhui Lin, Jingqun Tang, Hao Liu, Can Huang
| Challenge: | Current document image parsing solutions rely on specialized models or generate content autoregressively. |
| Approach: | They propose a multimodal document image parsing model that integrates specialized models with autogeneous content generation. |
| Outcome: | The proposed model achieves state-of-the-art performance across diverse page-level and element-level settings while ensuring superior efficiency. |
Agentic Oversight via Dialectic Reasoning (2026.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to align Large Language Models (LLMs) rely heavily on human annotations, but a Debate between expert models is a promising oversight mechanism. |
| Approach: | They propose a Debate between expert models to enable scalable oversight . they use a reasoning function to extend the framework to multilingual and multimodal spaces . |
| Outcome: | The proposed framework outperforms single-expert baselines in six multilingual and multimodal scenarios and shows that argument-mediated supervision instils unsupervised reasoning signals in expert models. |
MEXA: Towards General Multimodal Reasoning with Dynamic Multi-Expert Aggregation (2025.findings-emnlp)
Copied to clipboard
| Challenge: | MEXA is a training-free framework that performs modality- and task-aware aggregation of multiple expert models to enable effective multimodal reasoning across diverse domains. |
| Approach: | MEXA is a training-free framework that performs modality- and task-aware aggregation of multiple expert models. |
| Outcome: | MEXA performs modality- and task-aware aggregation of multiple expert models . it generates interpretable textual reasoning outputs and reasons over them using a Large Reasoning Model (LRM) MEX A consistently delivers performance improvements over strong multimodal benchmarks . |
Scaling Up Temporal Domain Generalization via Temporal Experts Averaging (2025.emnlp-main)
Copied to clipboard
Aoming Liu, Kevin Miller, Venkatesh Saligrama, Kate Saenko, Boqing Gong, Ser-Nam Lim, Bryan A. Plummer
| Challenge: | Temporal Domain Generalization (TDG) aims to generalize across temporal distribution shifts, e.g., lexical change over time. |
| Approach: | They propose a framework that updates the entire model using weight averaging to maximize generalization potential while minimizing computational costs. |
| Outcome: | The proposed framework outperforms previous methods by up to 69% while being up to 60x more efficient. |
Lightweight and Faithful Visual Condition Checking in Behavior Trees via Expert-Regularized Reinforcement Learning (2026.acl-long)
Copied to clipboard
| Challenge: | Existing behavior trees are not suitable for high-dimensional perceptual inputs such as images or language. |
| Approach: | They propose a framework that leverages expert-regularized reinforcement learning to preserve semantic faithfulness while employing a factorized policy that aggregates sequential condition-node decisions into a single decision unit. |
| Outcome: | The proposed framework outperforms imitation learning and reinforcement learning but risks misalignment of condition nodes with intended semantics and poor credit assignment. |
Verified Critical Step Optimization for LLM Agents (2026.findings-acl)
Copied to clipboard
| Challenge: | Critical Step Optimization (CSO) focuses preference learning on verified critical steps where alternative actions demonstrably flip task outcomes from failure to success. |
| Approach: | They propose a method which focuses preference learning on verified critical steps where alternative actions demonstrably flip task outcomes from failure to success. |
| Outcome: | The proposed method outperforms the existing methods on GAIA-Text-103 and XBench-DeepSearch while requiring supervision at only 16% of trajectory steps. |