Papers with knowledge distillation
IAG: Induction-Augmented Generation Framework for Answering Reasoning Questions (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to QA using retrieval-augmented knowledge are limited by limited coverage and noisy information. |
| Approach: | They propose an induction-augmented generation framework that utilizes inductive knowledge along with retrieved documents for implicit reasoning. |
| Outcome: | The proposed framework outperforms RAG and ChatGPT on two Open-Domain QA tasks. |
Infusing Sequential Information into Conditional Masked Translation Model with Self-Review Mechanism (2020.coling-main)
Copied to clipboard
| Challenge: | Existing non-autoregressive models generate target words in parallel, but with a large latency due to the left-to-right dependency. |
| Approach: | They propose to train a conditional masked translation model and refine results within several iterations to remedy a flawed translation by non-autoregressive models. |
| Outcome: | The proposed model outperforms state-of-the-art models by over 1 BLEU while using less training computations. |
Weight-Inherited Distillation for Task-Agnostic BERT Compression (2024.findings-naacl)
Copied to clipboard
| Challenge: | Knowledge Distillation (KD) is a predominant approach for BERT compression. |
| Approach: | They propose a weight-inherited distillation method which directly transfers knowledge from the teacher to a compact student model by inheriting the weights. |
| Outcome: | The proposed method outperforms state-of-the-art KD-based methods on GLUE and SQUAD benchmarks. |
EasyNLP: A Comprehensive and Easy-to-use Toolkit for Natural Language Processing (2022.emnlp-demos)
Copied to clipboard
Chengyu Wang, Minghui Qiu, Taolin Zhang, Tingting Liu, Lei Li, Jianing Wang, Ming Wang, Jun Huang, Wei Lin
| Challenge: | Pre-Trained Models (PTMs) have reshaped the development of natural language processing (NLP) but it is not easy to obtain high-performing PTMs without a large amount of labeled training data and deploy them online with fast inference speed. |
| Approach: | They propose to make it easy to build NLP applications with knowledge-enhanced pre-training and knowledge distillation. |
| Outcome: | EasyNLP supports a comprehensive suite of NLP algorithms and features knowledge-enhanced pre-training, knowledge distillation and few-shot learning functionalities. |
High Performance Natural Language Processing (2020.emnlp-tutorials)
Copied to clipboard
| Challenge: | a tutorial on scaling natural language processing will recapitulate the state-of-the-art in the field . |
| Approach: | This cutting-edge tutorial recapitulates the state-of-the-art in natural language processing with scale in perspective. |
| Outcome: | This cutting-edge tutorial recapitulates the state-of-the-art in natural language processing with scale in perspective. |
Efficient Transformer Knowledge Distillation: A Performance Review (2023.emnlp-industry)
Copied to clipboard
| Challenge: | Pretrained transformer language models have been gaining popularity in the field of natural language processing . however, there is no study into the intersection of these two fields . |
| Approach: | They propose a method to extract knowledge from transformers to produce high-performing efficient attention models with low costs. |
| Outcome: | The proposed model compression method preserves up to 98.6% of original model performance across short-context tasks and up to 95.8% on long-concept Named Entity Recognition tasks while decreasing inference times by up to 57%. |
Towards Non-task-specific Distillation of BERT via Sentence Representation Approximation (2020.aacl-main)
Copied to clipboard
| Challenge: | Existing methods for transferring knowledge from BERT into a model with large parameters are not efficient due to their large-scale and high computational cost. |
| Approach: | They propose a sentence representation approximating oriented distillation framework that can distill pre-trained BERT into a simple LSTM based model without specifying tasks. |
| Outcome: | The proposed model outperforms other distillation methods and larger models on multiple NLP tasks with efficiency well-improved. |
INDUS: Effective and Efficient Language Models for Scientific Applications (2024.emnlp-industry)
Copied to clipboard
Bishwaranjan Bhattacharjee, Aashka Trivedi, Masayasu Muraoka, Muthukumaran Ramasubramanian, Takuma Udagawa, Iksha Gurung, Nishan Pantha, Rong Zhang, Bharath Dandala, Rahul Ramachandran, Manil Maskey, Kaylin Bugbee, Michael Little, Elizabeth Fancher, Irina Gerasimov, Armin Mehrabian, Lauren Sanders, Sylvain Costes, Sergi Blanco-Cuaresma, Kelly Lockhart, Thomas Allen, Felix Grezes, Megan Ansdell, Alberto Accomazzi, Yousef El-Kurdi, Davis Wertheimer, Birgit Pfitzmann, Cesar Berrospi Ramis, Michele Dolfi, Rafael Lima, Panagiotis Vagenas, S. Mukkavilli, Peter Staar, Sanaz Vahidinia, Ryan McGranaghan, Tsengdar Lee
| Challenge: | Large language models trained on general domain corpora showed remarkable results on natural language processing tasks. |
| Approach: | They develop a suite of large language models trained on general domain corpora that address NLP tasks and smaller versions of them created using knowledge distillation. |
| Outcome: | The proposed models outperform general-purpose and domain-specific encoders on new and existing tasks and in industrial settings. |
Domain-specific transformer models for query translation (2023.acl-industry)
Copied to clipboard
| Challenge: | In domains such as Grocery, users prefer to buy certain brands of products . a large non-English speaking population makes it difficult to translate code-mix queries . |
| Approach: | They propose a model to preserve/correct Grocery brand names while translating context words . they propose to use a dataset of popular Groceries brand names to train the model . |
| Outcome: | The proposed model preserves/corrects Grocery brand names while translating context words . it is tested with a large non-English speaking population and is deployed in production . |
A Diverse and Effective Retrieval-Based Debt Collection System with Expert Knowledge (2025.naacl-industry)
Copied to clipboard
| Challenge: | Existing debt collection systems lack script diversity, contextual relevance and coherence due to their complexity. |
| Approach: | They propose a script library based on real-world debt collection conversations and a retrieval based response system for contextual relevance. |
| Outcome: | The proposed system improves script diversity and responds to debtor-collector conversations better through knowledge distillation. |
Meta-Learning Adaptive Knowledge Distillation for Efficient Biomedical Natural Language Processing (2022.findings-aacl)
Copied to clipboard
| Challenge: | Existing knowledge distillation methods have been proposed to reduce the size of large models for biomedical natural language processing tasks. |
| Approach: | They propose a meta-learning approach which adaptively learns parameters that enable optimal rate of knowledge exchange between teacher and student models from the distillation data during knowledge distillation. |
| Outcome: | The proposed method improves the performance of knowledge distillation methods on two biomedical natural language processing tasks. |
Beyond One-Step Distillation: Bridging the Capacity Gap in Small Language Models via Multi-Step Knowledge Transfer (2026.eacl-srw)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) excel across diverse tasks but remain too large for efficient on-device deployment. |
| Approach: | They revisit multi-step knowledge distillation as an effective remedy . they demonstrate that MSKD improves ROUGE-L and perplexity over single-step approaches . |
| Outcome: | The proposed approach improves ROUGE-L and perplexity over single-step approaches . large language models are too large for efficient on-device deployment, the authors show . |
Generation-Distillation for Efficient Natural Language Understanding in Low-Data Settings (D19-61)
Copied to clipboard
| Challenge: | Recent research points to knowledge distillation as a potential solution for NLU tasks. |
| Approach: | They propose a training approach that distills large finetuned LMs into a small network using unlabeled training examples. |
| Outcome: | The proposed approach outperforms BERT training approaches while using 300 times fewer parameters. |
A Study of Non-autoregressive Model for Sequence Generation (2020.acl-main)
Copied to clipboard
| Challenge: | Non-autoregressive (NAR) models generate all tokens in parallel, resulting in faster generation speed compared to autoregressive models. |
| Approach: | They propose to use knowledge distillation and source-target alignment to bridge the gap between NAR and autoregressive models in various tasks. |
| Outcome: | The proposed techniques can speed up NAR models in some tasks but not all . the proposed techniques reduce target token dependency while allowing for faster inference . |
GKD: A General Knowledge Distillation Framework for Large-scale Pre-trained Language Model (2023.acl-industry)
Copied to clipboard
| Challenge: | Existing knowledge distillation frameworks for language models are limited by memory and the use of complex distillation methods on larger-scale PLMs. |
| Approach: | They propose a general knowledge distillation framework that supports distillation on larger-scale PLMs using various distillation methods. |
| Outcome: | The proposed framework can support distillation on larger-scale PLMs and 25 mainstream methods on 8 NVIDIA A100 (40GB) GPUs. |
Resource-Efficient Anonymization of Textual Data via Knowledge Distillation from Large Language Models (2025.coling-industry)
Copied to clipboard
| Challenge: | Existing approaches to anonymize textual data from large language models pose privacy risks due to their API-based access. |
| Approach: | They propose a method to distill large language models into smaller encoder-only models via named entity recognition coupled with regular expressions to create a lightweight model capable of effective anonymization. |
| Outcome: | The proposed approach reduces computational overhead while maintaining semantic integrity of data. |
TAKE: Topic-shift Aware Knowledge sElection for Dialogue Generation (2022.coling-1)
Copied to clipboard
| Challenge: | Recent work finds that realizing who holds the initiative can help select knowledge . however, there is a strong semantic transition between two rounds, probably leading to initiative misjudgment . |
| Approach: | They propose a topic-shift Aware Knowledge sElector(TAKE) model which locates relevant parts from dialogue history to improve knowledge selection. |
| Outcome: | The proposed model outperforms baseline models on the WoW. |
Cost-effective Deployment of BERT Models in Serverless Environment (2021.naacl-industry)
Copied to clipboard
| Challenge: | a large upfront infrastructure investment makes machine learning models difficult to deploy . however, serverless architectures have strict limits on the size of the deployment package . |
| Approach: | They propose to fine-tune BERT-style models on proprietary datasets for tasks . they use knowledge distillation to obtain models that are tuned for a specific domain . |
| Outcome: | The proposed model deployments report acceptable latency levels and cost-effectiveness without infrastructure overhead. |
The economic trade-offs of large language models: A case study (2023.acl-industry)
Copied to clipboard
Kristen Howell, Gwen Christian, Pavel Fomitchov, Gitit Kehat, Julianne Marzulla, Leanne Rolston, Jadin Tredup, Ilana Zimmerman, Ethan Selfridge, Joseph Bradley
| Challenge: | Large Language Models (LLMs) are a natural fit for contact-based customer service, but their efficacy must be balanced with the cost of training and serving them. |
| Approach: | They propose a cost framework for evaluating an NLP model’s utility for the enterprise as a function of the usefulness of the responses that they generate. |
| Outcome: | The proposed model can be used to help human agents handle complex customer service calls and can be modified to improve their performance. |
Benchmarking Diffusion Models for Machine Translation (2024.eacl-srw)
Copied to clipboard
| Challenge: | Diffusion models have shown great potential on many generative tasks, but their application to natural language processing (NLP) is still a less explored direction. |
| Approach: | They adapt two diffusion-based text generation models, Diffusion-LM and DiffuSeq, to perform machine translation. |
| Outcome: | The proposed models struggle more on long-range dependencies than other models. |
When Speed Meets Intelligence: Scalable Conversational NER in an Ever-evolving World (2026.eacl-industry)
Copied to clipboard
| Challenge: | Large Language Models excel at understanding conversational semantics, but lack of data makes them impractical for production deployment. |
| Approach: | They propose a pipeline for generating multilingual conversational NER datasets with minimal human validation and a framework that leverages LLMs as semantic filters combined with catalog-based entity grounding to label live traffic data. |
| Outcome: | The proposed framework outperforms existing models on public and private conversations by 97.12% on CoNLL-2003 and 83.09% on OntoNotes 5.0. |
Sample, Translate, Recombine: Leveraging Audio Alignments for Data Augmentation in End-to-end Speech Translation (2022.acl-short)
Copied to clipboard
| Challenge: | End-to-end speech translation relies on data that pair source-language speech inputs with corresponding translations. |
| Approach: | They propose a method that augments transcriptions by sampling from suffix memory and translating them into target languages. |
| Outcome: | The proposed method delivers up to 0.9 and 1.1 BLEU points on top of augmentation with knowledge distillation on languages on CoVoST 2 and Europarl-ST. |
Performance-Efficiency Trade-Offs in Adapting Language Models to Text Classification Tasks (2022.aacl-short)
Copied to clipboard
| Challenge: | Pre-trained language models (LMs) are state-of-the-art when adapted to text classification tasks. |
| Approach: | They compare fine-tuning, prompting, and knowledge distillation procedures to train pre-trained language models to downstream tasks. |
| Outcome: | The proposed training procedures perform better when trained with fine-tuning or prompting on large train sets than when trained by prompting or fine-untun. |
Distilling the Knowledge of Romanian BERTs Using Multiple Teachers (2022.lrec-1)
Copied to clipboard
Andrei-Marius Avram, Darius Catrina, Dumitru-Clementin Cercel, Mihai Dascalu, Traian Rebedea, Vasile Pais, Dan Tufis
| Challenge: | Existing approaches to train pre-trained language models focus on the English language, thus widening the gap when considering low-resource languages. |
| Approach: | They propose three versions of distilled BERT models for the Romanian language . they argue that the models offer performance comparable to their teachers . |
| Outcome: | The proposed models perform comparable to their teachers, while being twice as fast on a GPU and 35% smaller. |
PRAL: A Tailored Pre-Training Model for Task-Oriented Dialog Generation (2021.acl-short)
Copied to clipboard
| Challenge: | Existing approaches to building task-oriented dialog systems require a substantial amount of annotations and thus are labor-intensive. |
| Approach: | They propose a Pre-trainedRole Alternating Language model (PRAL) that is explicitly designed for task-oriented dialog tasks. |
| Outcome: | The proposed model outperforms or is on par with state-of-the-art models on task-oriented dialog tasks. |
Pre-Training Methods for Question Reranking (2024.eacl-short)
Copied to clipboard
| Challenge: | Existing methods for Question Answering to search for semantically similar questions are not suitable for new questions. |
| Approach: | They propose an unsupervised method for retrieving and ranking questions . they use a question retrieval model and a selection model to rerank questions based on their relevance . |
| Outcome: | The proposed method achieves state-of-the-art performance on QRC and Quora-match datasets . it provides better and cheaper access to answers than the system generated them . |
deepQuest-py: Large and Distilled Models for Quality Estimation (2021.emnlp-demo)
Copied to clipboard
Fernando Alva-Manchego, Abiola Obamuyide, Amit Gajbhiye, Frédéric Blain, Marina Fomicheva, Lucia Specia
| Challenge: | Quality Estimation (QE) is a tool for machine translation that predicts how good translations are without comparing them to gold-standard references. |
| Approach: | They introduce a framework for training and evaluation of large and light-weight models for Quality Estimation (QE) they use pre-trained Transformers to train large and efficient QE models. |
| Outcome: | The framework provides access to state-of-the-art models based on pre-trained Transformers for sentence-level and word-level QE and a web interface for testing and visualising their predictions. |
Streamlining LLMs: Adaptive Knowledge Distillation for Tailored Language Models (2025.naacl-srw)
Copied to clipboard
| Challenge: | Large language models (LLMs) have transformative potential across industries, e.g., enhancing customer service, revolutionizing medical diagnostics, or identifying crises in news articles. |
| Approach: | They propose to distill compact, parameter-efficient tailored language models from LLMs for domain-specific tasks with comparable performance. |
| Outcome: | The proposed framework outperforms knowledge distillation frameworks in the crisis domain, where labeled data is scarce. |
Incremental Sequence Labeling: A Tale of Two Shifts (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to incremental sequence labeling have focused on the E2O and O2E issues, but neglect the O2e issue. |
| Approach: | They propose a framework for incremental sequence labeling without semantic shifts that mitigate catastrophic forgetting in models by using knowledge distillation to maintain the model’s discriminative ability for old entities. |
| Outcome: | The proposed framework mitigates catastrophic forgetting in models while maintaining discriminative ability for old entities while minimizing the model’s bias towards new entities. |
Generate, Annotate, and Learn: NLP with Synthetic Text (2022.tacl-1)
Copied to clipboard
| Challenge: | Existing methods to generate unlabeled text are difficult to find. |
| Approach: | They propose a general framework called "generate, annotate, and learn" to take advantage of synthetic text within knowledge distillation, self-training, and few-shot learning applications. |
| Outcome: | The proposed framework achieves state-of-the-art knowledge distillation results for 6-layer transformers on the GLUE leaderboard. |
Speed Without Sacrifice: Fine-Tuning Language Models with Medusa and Knowledge Distillation in Travel Applications (2025.acl-industry)
Copied to clipboard
Daniel Zagyva, Emmanouil Stergiadis, Laurens Van Der Maas, Aleksandra Dokic, Eran Fainman, Ilya Gusev, Moran Beladev
| Challenge: | Rapid growth of digital applications has intensified the demand for real-time natural language processing (NLP) capabilities. |
| Approach: | They propose a framework that combines Medusa and knowledge distillation to achieve compounded benefits in both model size and inference speed. |
| Outcome: | The proposed framework reduces inference latency by 10-20x while maintaining the student model’s performance quality. |
QUILL: Query Intent with Large Language Models using Retrieval Augmentation and Multi-stage Distillation (2022.emnlp-industry)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have shown impressive results on a variety of text understanding tasks. |
| Approach: | They propose a two-stage distillation approach that allows retrieval augmentation to be carried over without the increased compute associated with it. |
| Outcome: | The proposed approach can carry over the gains of retrieval augmentation without suffering the increased compute typically associated with it. |
On the Use of External Data for Spoken Named Entity Recognition (2022.naacl-main)
Copied to clipboard
| Challenge: | Named entity recognition (NER) tasks require large labeled datasets to perform . compared to prior work, relative improvements in F1 of up to 16% are found . |
| Approach: | They propose to use self-training, knowledge distillation, and transfer learning to learn SLU models . they compare pipeline and pipeline approaches to find out how to use external data . |
| Outcome: | The proposed models improve performance beyond pre-trained models in resource-constrained settings . the best baseline model is a pipeline approach, while the best performance is achieved by an E2E model. |
Mitigating Gender Bias in Distilled Language Models via Counterfactual Role Reversal (2022.findings-acl)
Copied to clipboard
Umang Gupta, Jwala Dhamala, Varun Kumar, Apurv Verma, Yada Pruksachatkun, Satyapriya Krishna, Rahul Gupta, Kai-Wei Chang, Greg Ver Steeg, Aram Galstyan
| Challenge: | Language models excel at generating coherent text, but can be biased in multiple ways, including the unfounded association of male and female genders with gender-neutral professions. |
| Approach: | They propose to modify teacher probabilities and augment the training set to learn a fair model during knowledge distillation by modifying teacher probability and augmenting the training sets. |
| Outcome: | The proposed approach reduces gender disparity in open-ended text generated from the distilled and finetuned models with only a minor compromise in utility. |
When Gradient Descent Meets Derivative-Free Optimization: A Match Made in Black-Box Scenario (2023.findings-acl)
Copied to clipboard
| Challenge: | Large pre-trained language models (PLMs) are expensive and may not be open-sourced due to commercial considerations and potential risks of misuse. |
| Approach: | They propose to introduce gradient descent into black-box tuning scenario . they propose a method which integrates gradient descent and derivative-free optimization . |
| Outcome: | The proposed method achieves significant performance gains over previous state-of-the-art methods. |
Plug and Play Knowledge Distillation for kNN-LM with External Logits (2022.aacl-short)
Copied to clipboard
| Challenge: | Despite the promising evaluation results by knowledge distillation (KD) in natural language understanding (NLU) and sequence-to-sequence (seq2sequ) tasks, KD for causal language modeling (LM) remains a challenge. |
| Approach: | They propose to use external logits to improve a student's kNN-LM by leveraging teacher's knowledge at test time. |
| Outcome: | The proposed method improves a student's kNN-LM in multiple language modeling datasets and improves perplexity. |
DisComp: A Two-Stage Prompt Optimization Framework Combining Task-Agnostic and Task-Aware Compression (2025.findings-naacl)
Copied to clipboard
| Challenge: | Extended prompts can lead to substantial computational overhead and increased hardware demands, limiting the scalability and efficiency of large language models. |
| Approach: | They propose a two-stage prompt compression framework that combines task-agnostic and task-based strategies to efficiently compress prompt length without compromising performance. |
| Outcome: | The proposed framework outperforms task-agnostic and task-specific compression methods on three benchmark datasets and is up to 6.56 faster at inference compared to the best token-level compression method. |
PILE: Pairwise Iterative Logits Ensemble for Multi-Teacher Labeled Distillation (2022.emnlp-industry)
Copied to clipboard
Lianshang Cai, Linhao Zhang, Dehong Ma, Jun Fan, Daiting Shi, Yi Wu, Zhicong Cheng, Simiu Gu, Dawei Yin
| Challenge: | Pre-trained language models have been a key part of ranking systems . knowledge distillation is widely used to maintain high performance while keeping efficient computations. |
| Approach: | They propose an algorithm to combine knowledge from multi-teachers and label information to achieve competitive performance in offline and online experiments. |
| Outcome: | The proposed method has been deployed in a real-world commercial search system. |
AMR-Evol: Adaptive Modular Response Evolution Elicits Better Knowledge Distillation for Large Language Models in Code Generation (2024.emnlp-main)
Copied to clipboard
| Challenge: | proprietary large language models (LLMs) have demonstrated impressive code generation performance. |
| Approach: | They propose an adaptive module-based model that refines the direct response distillation process by modular decomposition and adaptive response evolution. |
| Outcome: | The proposed framework outperforms baseline model and code generation methods on three popular benchmarks. |
CUPID: Curriculum Learning Based Real-Time Prediction using Distillation (2023.acl-industry)
Copied to clipboard
| Challenge: | Relevance in E-commerce Product Search is crucial for providing customers with accurate results that match their query intent. |
| Approach: | They propose a curriculum learning based real-time relevance prediction using distillation . they propose e-commerce search systems that use transformers to predict relevance . |
| Outcome: | The proposed model improves on english and Arabic in a bi-lingual relevance prediction task while maintaining low evaluation latency on CPUs. |
Why Skip If You Can Combine: A Simple Knowledge Distillation Technique for Intermediate Layers (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing knowledge distillation techniques are not suitable for deep learning tasks due to memory constraints. |
| Approach: | They propose to combine knowledge from a large teacher network into a student network (S) they propose to use a combinatorial mechanism to inject layer-level supervision from T to S . |
| Outcome: | The proposed model outperforms existing models in PortugueseEnglish, TurkishEnglish and EnglishGerman directions and students trained using it have 50% fewer parameters and can deliver comparable results to 12-layer teachers. |
Iterative Dual Domain Adaptation for Neural Machine Translation (D19-1)
Copied to clipboard
| Challenge: | Empirical results on Chinese-English and English-German translation tasks demonstrate the effectiveness of our proposed framework. |
| Approach: | They propose an iterative dual domain adaptation framework for neural machine translation that uses multiple corpora to perform bidirectional translation knowledge transfer. |
| Outcome: | Empirical results on Chinese-English and English-German translation tasks demonstrate the effectiveness of the proposed framework. |
Flexible Weight Tuning and Weight Fusion Strategies for Continual Named Entity Recognition (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for Named Entity Recognition (CNER) use knowledge distillation to retain old knowledge, but they are too expensive and fail to integrate with existing state-of-the-art models. |
| Approach: | They propose a weight tuning and weightfusion strategy to learn new entity types while mitigating catastrophic forgetting of old models. |
| Outcome: | The proposed strategies improve the performance of existing models and are model-agnostic. |
Bootstrapping meaning through listening: Unsupervised learning of spoken sentence embeddings (2022.findings-emnlp)
Copied to clipboard
| Challenge: | a new study addresses the challenge of learning semantic representations from speech signals . speech-based semantic representation can be used for speech mining and spoken language understanding . |
| Approach: | They propose a multimodal sequential autoencoder that converts speech signals into hidden units . they propose s-HuBERT to induce meaning through knowledge distillation . |
| Outcome: | The proposed model achieves a moderate correlation with human judgments without labels or transcriptions. |
When Chosen Wisely, More Data Is What You Need: A Universal Sample-Efficient Strategy For Data Augmentation (2022.findings-acl)
Copied to clipboard
| Challenge: | Existing DA methods naively add a certain number of augmented samples without considering the quality and the added computational cost of these samples. |
| Approach: | They propose a data-augmented DA technique that generates or reweights augmented samples . they say it is faster to train and can be plugged into any DA method . |
| Outcome: | The proposed technique is faster to train and more efficient than existing methods. |
MATE-KD: Masked Adversarial TExt, a Companion to Knowledge Distillation (2021.acl-long)
Copied to clipboard
| Challenge: | Recent studies have shown that the trillion parameter mark for pre-trained language models is not achievable without a change in training paradigm. |
| Approach: | They propose a text-based adversarial training algorithm which enhances the performance of knowledge distillation by maximizing the divergence between teacher and student logits. |
| Outcome: | The proposed algorithm outperforms competing adversarial learning and data augmentation baselines on the GLUE benchmark. |
A Flexible Multi-Task Model for BERT Serving (2022.acl-short)
Copied to clipboard
| Challenge: | a proposed BERT-based multi-task framework is suitable for iterative and incremental development of the tasks. |
| Approach: | They propose an efficient BERT-based multi-task framework that is suitable for iterative and incremental development of the tasks. |
| Outcome: | The proposed framework achieves 99.6% of performance of the full fine-tuning method while reducing up to two thirds of its overhead. |
COSIGN: Contextual Facts Guided Generation for Knowledge Graph Completion (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing methods for knowledge graph completion (KGC) are limited in generality and scalability due to poor contextual facts. |
| Approach: | They propose a contextual facts collector and contextual facts organizer to enhance the inference ability of GM-based methods for various KGC tasks. |
| Outcome: | The proposed model outperforms state-of-the-art methods in terms of performance. |
Integrating Translation Memories into Non-Autoregressive Machine Translation (2023.eacl-main)
Copied to clipboard
| Challenge: | Non-autoregressive machine translation (NAT) has made great progress, but most studies focus on standard translation tasks. |
| Approach: | They propose to train an edit-based NAT model with a Translation Memory (TM) they propose to modify the data presentation and introduce an extra deletion operation to reduce decoding load. |
| Outcome: | The proposed model performs on par with an autoregressive approach while reducing the decoding load. |
QABISAR: Query-Article Bipartite Interactions for Statutory Article Retrieval (2025.coling-main)
Copied to clipboard
| Challenge: | Existing methods for Statutory Article Retrieval (SAR) are vague and underspecified . however, a new approach is needed to bridge the gap between legal expertise and public understanding . |
| Approach: | They propose a framework for statutory article retrieval that leverages bipartite interactions between queries and articles to capture diverse aspects inherent in them. |
| Outcome: | The proposed framework overcomes the semantic mismatch problem when modeling each query-article pair in isolation. |
Unsupervised Knowledge Selection for Dialogue Generation (2021.findings-acl)
Copied to clipboard
| Challenge: | Existing knowledge selection tasks require the preidentified knowledge to generate informative dialogues. |
| Approach: | They propose a novel method to supervise knowledge selection when the gold knowledge label is unknown by obtaining an oracle knowledge label via distant supervision and leverage knowledge distillation to alleviate the noisy labeling problem of distant supervision. |
| Outcome: | The proposed method outperforms strong supervised baselines on two knowledge-grounded dialogue datasets and generates more informative responses. |
Learning to Retrieve In-Context Examples for Large Language Models (2024.eacl-long)
Copied to clipboard
| Challenge: | Existing approaches to improve in-context learning performance are highly sensitive to the quality of the incontext examples provided. |
| Approach: | They propose a framework to iteratively train dense retrievers that can identify high-quality in-context examples for large language models. |
| Outcome: | The proposed model improves performance by retrieving examples with similar patterns, and the gains are consistent across LLMs of varying sizes. |
Combining Curriculum Learning and Knowledge Distillation for Dialogue Generation (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Existing studies have shown that curriculum learning facilitates dialogue generation tasks while knowledge distillation can yield significant performance boosts for student models. |
| Approach: | They propose a combination of curriculum learning and knowledge distillation for dialogue generation models . they cluster training cases according to their complexity and employ an adversarial training strategy . |
| Outcome: | The proposed model improves compared with baselines. |
MoEBERT: from BERT to Mixture-of-Experts via Importance-Guided Adaptation (2022.naacl-main)
Copied to clipboard
| Challenge: | Existing methods for training pre-trained language models have limited practicality due to latency requirements. |
| Approach: | They propose a method that uses a Mixture-of-Experts structure to increase model capacity and inference speed. |
| Outcome: | The proposed method outperforms existing distillation methods on natural language understanding and question answering tasks. |
Domain Knowledge Transferring for Pre-trained Language Model via Calibrated Activation Boundary Distillation (2022.acl-long)
Copied to clipboard
| Challenge: | Pretrained language models are used to boost their performance on downstream tasks . pretraining with in-domain texts requires considerable in- domain data and training resources . |
| Approach: | They propose a domain knowledge transferring framework for pre-trained language models without additional in-domain pretraining. |
| Outcome: | The proposed framework extracts domain knowledge from an existing in-domain pretrained language model and transfers it to other PLMs by applying knowledge distillation. |
GOVERN: Gradient Orientation Vote Ensemble for Multi-Teacher Reinforced Distillation (2024.emnlp-industry)
Copied to clipboard
| Challenge: | Pre-trained language models have achieved remarkable performance in OpenQA, but for practical deployment, knowledge distillation is crucial to maintain high performance while operating under computational constraints. |
| Approach: | They propose an algorithm to perform unsupervised knowledge distillation without the guidance of labels to achieve 99.5% of performance. |
| Outcome: | The proposed algorithm achieves 99.5% of performance in a commercial question-answering system. |
Robustness Challenges in Model Distillation and Pruning for Natural Language Understanding (2023.eacl-main)
Copied to clipboard
| Challenge: | Recent studies have focused on compressing pre-trained language models (PLMs) however, few studies have examined the impact of compression on generalizability and robustness of compressed models for out-of-distribution data. |
| Approach: | They propose to use knowledge distillation and pruning to reduce model generalization and generalization on out-of-distribution data. |
| Outcome: | The proposed compression techniques overfit on shortcut samples and generalize poorly on hard ones. |
Document-Level Relation Extraction with Adaptive Focal Loss and Knowledge Distillation (2022.findings-acl)
Copied to clipboard
| Challenge: | Document-level relation extraction (DocRE) is a more challenging task than sentence-level one. |
| Approach: | They propose a semi-supervised framework for document-level relation extraction with three components . they use an axial attention module for learning the interdependency among entity-pairs . |
| Outcome: | The proposed model outperforms baseline models on two DocRE datasets and outperformed previous models on human annotated data and distantly supervised data. |
Improving Neural Topic Models using Knowledge Distillation (2020.emnlp-main)
Copied to clipboard
| Challenge: | Current paradigms for transfer learning use general knowledge as a foundation for more specialized endeavors. |
| Approach: | They propose to combine probabilistic topic models and pretrained transformers to improve topic quality by using knowledge distillation. |
| Outcome: | The proposed framework improves topic quality over all estimated topics and in head-to-head comparisons of aligned topics. |
A Study on the Efficiency and Generalization of Light Hybrid Retrievers (2023.acl-short)
Copied to clipboard
Man Luo, Shashank Jain, Anchit Gupta, Arash Einolghozati, Barlas Oguz, Debojeet Chatterjee, Xilun Chen, Chitta Baral, Peyman Heidari
| Challenge: | Recent research focuses on building neural retrievers which learn dense embeddings of query and document into a semantic space. |
| Approach: | They propose to use an indexing-efficient dense retriever to reduce hybrid retrievers' memory by using the state-based indexing algorithm. |
| Outcome: | The proposed hybrid retriever saves 13 memory while maintaining 98.0% performance on out-of-domain datasets and adversarial attacks datasets. |
Teaching Small Language Models to Reason (2023.acl-short)
Copied to clipboard
| Challenge: | Chain of thought prompting improves reasoning capabilities of large language models but only emerges in models with tens of billions of parameters. |
| Approach: | They propose to fine tune a student model on chain of thought outputs generated by a larger teacher model. |
| Outcome: | The proposed method improves task performance across arithmetic, commonsense and symbolic reasoning datasets. |
A Simple Concatenation can Effectively Improve Speech Translation (2023.acl-short)
Copied to clipboard
| Challenge: | Experimental results show that in our unified cross-modal ST model, models can effectively utilize the auxiliary information from speech and text. |
| Approach: | They propose a unified cross-modal ST method which concatenates speech and text as the input and builds a teacher that can utilize both cross-modities simultaneously. |
| Outcome: | The proposed method can effectively utilize the auxiliary information from speech and text, and achieve compelling results on MuST-C datasets. |
Gradient-based Intra-attention Pruning on Pre-trained Language Models (2023.acl-long)
Copied to clipboard
| Challenge: | Pre-trained language models are computationally expensive and slow in inference due to their large sizes. |
| Approach: | They propose a structured pruning method which combines pruning with knowledge distillation to yield highly effective models. |
| Outcome: | The proposed method outperforms other pruning methods in sparsity regimes while maintaining 93% 99% performance. |
Weight Distillation: Transferring the Knowledge in Neural Network Parameters (2021.acl-long)
Copied to clipboard
| Challenge: | Knowledge distillation is an effective method for model acceleration and compression. |
| Approach: | They propose to use parameters to distill knowledge from large neural networks to small ones . they propose to do this by using a parameter generator to transfer the knowledge to a small neural network . |
| Outcome: | The proposed method learns a small network 1.88 2.94x faster than the large network but with competitive BLEU points. |
Boosting Summarization with Normalizing Flows and Aggressive Training (2023.emnlp-main)
Copied to clipboard
| Challenge: | Experimental results show that FlowSUM improves the quality of generated summaries with minimal impact on inference time. |
| Approach: | They propose a normalizing flows-based variational encoder-decoder framework for Transformer-based summarization. |
| Outcome: | The proposed model improves the quality of generated summaries and reduces inference time. |
FedED: Federated Learning via Ensemble Distillation for Medical Relation Extraction (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing relation extraction methods require centralizing training data from different medical platforms while holding the privacy-sensitive data puts patients' privacy at risk. |
| Approach: | They propose a federated relation extraction model that trains a central model without sharing or exchange of private local data. |
| Outcome: | The proposed model trains a central model without uploading local parameters, and it performs well on three publicly available datasets. |
UEGP: Unified Expert-Guided Pre-training for Knowledge Rekindle (2024.findings-naacl)
Copied to clipboard
Yutao Mou, Kexiang Wang, Jianhe Lin, Dehong Ma, Jun Fan, Daiting Shi, Zhicong Cheng, Gu Simiu, Dawei Yin, Weiran Xu
| Challenge: | Existing paradigms for pre-training and fine-tuning have limitations . knowledge rekindle aims to break through performance upper bounds of experts without introducing additional annotated data. |
| Approach: | They propose a new paradigm for pre-training and fine-tuning that aims to re-incorporate the fine- tuned expert model into the training cycle and break through performance upper bounds of experts. |
| Outcome: | The proposed model breaks through performance upper bounds of experts without additional annotated data. |
Improving Non-autoregressive Neural Machine Translation with Monolingual Data (2020.acl-main)
Copied to clipboard
| Challenge: | Neural machine translation is usually done via knowledge distillation from an autoregressive (AR) model. |
| Approach: | They leverage large monolingual corpora to improve the NAR model's performance by transferring the autoregressive model' s generalization ability while preventing overfitting. |
| Outcome: | The proposed methods on the WMT14 En-De and WMT16 En-Ro news translation tasks show that monolingual data augmentation improves the NAR model to approach the teacher AR model’s performance. |
PAIR: Leveraging Passage-Centric Similarity Relation for Improving Dense Passage Retrieval (2021.findings-acl)
Copied to clipboard
Ruiyang Ren, Shangwen Lv, Yingqi Qu, Jing Liu, Wayne Xin Zhao, QiaoQiao She, Hua Wu, Haifeng Wang, Ji-Rong Wen
| Challenge: | Recent studies only consider query-centric similarity relation when learning the dual-encoder retriever. |
| Approach: | They propose a query-centric and PAssage-centric approach to capture more comprehensive similarity relations for dense passage retrieval. |
| Outcome: | The proposed approach significantly outperforms existing models on both MSMARCO and Natural Questions datasets. |
Query Optimization for Parametric Knowledge Refinement in Retrieval-Augmented Large Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Extract-Refine-Retrieve-Read is a query optimization framework for large language models . it is designed to bridge the pre-retrieval information gap in Retriev-Augmented Generation systems . |
| Approach: | They propose a framework to extract parametric knowledge from Large Language Models and refine them using a specialized query optimizer. |
| Outcome: | The extract-refine-retrieve-read framework outperforms baselines on QA datasets . it is designed to meet the knowledge requirements of large language models (LLMs) |
Re2G: Retrieve, Rerank, Generate (2022.naacl-main)
Copied to clipboard
Michael Glass, Gaetano Rossiello, Md Faisal Mahbub Chowdhury, Ankita Naik, Pengshan Cai, Alfio Gliozzo
| Challenge: | Recent models such as RAG and REALM incorporate retrieval into conditional generation. |
| Approach: | They propose a method that combines retrieval and reranking into a BART-based sequence-to-sequence generation. |
| Outcome: | The proposed model combines retrieval and reranking into a BART-based sequence-to-sequence generation. |
CKDST: Comprehensively and Effectively Distill Knowledge from Machine Translation to End-to-End Speech Translation (2023.findings-acl)
Copied to clipboard
| Challenge: | End-to-end speech-totext translation (ST) data are limited due to the limited resources. |
| Approach: | They propose a knowledge distillation framework for speech translation that integrates knowledge from machine translation and decouples knowledge from non-target class knowledge. |
| Outcome: | The proposed framework outperforms state-of-the-art models on a benchmark dataset. |
XtremeDistil: Multi-stage Distillation for Massive Multilingual Models (2020.acl-main)
Copied to clipboard
| Challenge: | Existing work on pre-trained language models focuses on reducing the size of these models into shallow ones. |
| Approach: | They propose a knowledge distillation technique that leverages teacher internal representations to reduce the size of pre-trained language models. |
| Outcome: | The proposed method outperforms previous methods in multilingual Named Entity Recognition (NER) it reduces the size of teacher models by 35x while retaining 95% of its F1 score. |
Empowering Dual-Encoder with Query Generator for Cross-Lingual Dense Retrieval (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods to distill knowledge from cross-encoder re-ranker to dual-encoding retriever are lacking in the cross-lingual setting. |
| Approach: | They propose to use a query generator as the teacher in the cross-lingual setting to distill knowledge to a dual-encoder retrieval model. |
| Outcome: | The proposed method outperforms state-of-the-art methods on two benchmark datasets. |
GKT: A Novel Guidance-Based Knowledge Transfer Framework For Efficient Cloud-edge Collaboration LLM Deployment (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing methods of acceleration require fine-tuning of considerably large models, such as Llama-7B, posing a challenge for average users. |
| Approach: | They propose a Guidance-based Knowledge Transfer framework that leverages a larger LLM as a 'teacher' and a smaller 'student' model to finalize responses. |
| Outcome: | The proposed framework achieves a maximum accuracy improvement of 14.18%, along with a 10.72 times speed-up on GSM8K and an accuracy improvement 14.00% along with 7.73 times speed up in CSQA. |
Confidence Based Bidirectional Global Context Aware Training Framework for Neural Machine Translation (2022.acl-long)
Copied to clipboard
| Challenge: | Existing studies focus on how to effectively exploit bidirectional global contexts in neural machine translation models. |
| Approach: | They propose a Confidence Based Bidirectional Global Context Aware training framework for NMT . they incorporate bidirectional global context to the NMT model on unconfidently-predicted target words . |
| Outcome: | The proposed framework improves the NMT model on three large-scale translation datasets by +1.02, +0.57 BLEU scores. |
Attribute-Controlled Translation with Preference Optimization (2026.findings-eacl)
Copied to clipboard
| Challenge: | Attribute-controlled translation (ACT) is a natural language processing task that produces translations that satisfy specific constraints on linguistic and stylistic attributes. |
| Approach: | They propose to leverage the contrastive nature of ACT tasks with preference optimization . they also propose to exploit knowledge distillation with synthetically-generated training samples . |
| Outcome: | The proposed approach improves attribute matching and translation quality in small-medium size models. |
Towards Higher Pareto Frontier in Multilingual Machine Translation (2023.acl-long)
Copied to clipboard
| Challenge: | Existing Pareto optimization approaches are limited by the long-tailed distribution of multilingual corpora. |
| Approach: | They propose a Pareto mutual distillation framework that pushes the Paret frontier outwards rather than making trade-offs. |
| Outcome: | The proposed framework pushes the Pareto frontier outwards rather than making trade-offs, the authors show. |
Annealing Knowledge Distillation (2021.eacl-main)
Copied to clipboard
| Challenge: | Knowledge distillation (KD) is a powerful model compression technique for deep neural networks. |
| Approach: | They propose a method to feed the rich information provided by teacher’s soft-targets incrementally and more efficiently by annealing the teacher output incrementally. |
| Outcome: | The proposed method can be used on image classification and NLP language inference tasks with BERT-based models on the GLUE benchmark. |
Normalizing without Modernizing: Keeping Historical Wordforms of Middle French while Reducing Spelling Variants (2024.findings-naacl)
Copied to clipboard
| Challenge: | a new method to normalize orthographic variations of historical documents is needed for digital humanities and diachronic studies. |
| Approach: | They propose to normalize orthographic wordforms found in Middle French archives . authors say it improves accuracy and accuracy over a strong baseline . |
| Outcome: | The proposed methods normalize orthographic variations of historical documents without modernizing them. |
Learning from Committee: Reasoning Distillation from a Mixture of Teachers with Peer-Review (2025.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) have proven to be highly effective in addressing a wide range of complex tasks. |
| Approach: | They propose a method that asks teachers to identify and explain student’s mistakes and then asks them to provide customized instruction learning data. |
| Outcome: | The proposed method reduces the chance of teachers guessing incorrectly with flawed rationales, improving instructional data quality. |
Bridging Cognition and Affect: Emotion-Aware Opinion Summarization using LLMs (2026.findings-acl)
Copied to clipboard
| Challenge: | Emotion-aware Opinion Summarization (EAOS) is a framework that captures emotions that shape purchasing decisions. |
| Approach: | They propose a framework that integrates emotion into opinion summaries and a large-scale training dataset and an evaluation benchmark to support this task. |
| Outcome: | The proposed framework captures discrete emotions that shape purchasing decisions. |
Guiding Teacher Forcing with Seer Forcing for Neural Machine Translation (2021.acl-long)
Copied to clipboard
| Challenge: | Neural machine translation models are usually based on attention-based encoder-decoder frameworks. |
| Approach: | They introduce a seer decoder into the encoder-decoder framework during training . they force the conventional decoded decodes to simulate the behavior of the seer . |
| Outcome: | The proposed method outperforms baselines on Chinese, English and German translation tasks. |
AlignCap: Aligning Speech Emotion Captioning to Human Preferences (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for speech emotion capture often produce hallucinations and lose generalization on unseen speech. |
| Approach: | They propose to align speech emotion captioning to human preference based on large language model (LLM) and human preference regularization to eliminate factuality and faithfulness hallucinations. |
| Outcome: | Experiments show that AlignCap performs better than existing methods on Zero-shot SEC task. |
Marginal Utility Diminishes: Exploring the Minimum Knowledge for BERT Knowledge Distillation (2021.acl-long)
Copied to clipboard
| Challenge: | Knowledge distillation (KD) has shown great success in BERT compression. |
| Approach: | They propose a knowledge distillation paradigm that extracts the teacher's hidden state knowledge and then compresses it into three dimensions. |
| Outcome: | The proposed paradigm gives rise to training speedup of 2.7x 3.4x for two kinds of student models and computing devices. |
Attention-Guided Answer Distillation for Machine Reading Comprehension (D18-1)
Copied to clipboard
| Challenge: | Existing approaches to reading comprehension systems are vulnerable to adversarial attacks. |
| Approach: | They propose to use knowledge distillation to transfer knowledge from an ensemble to a single model. |
| Outcome: | The proposed methods outperform the teacher on adversarial datasets and NarrativeQA benchmarks. |
Lifelong Language Knowledge Distillation (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods to perform lifelong language learning (LLL) on stream of different tasks are challenging . Existing models face catastrophic forgetting problem, which can be mitigated by lifelong learning . |
| Approach: | They propose a method that can be easily applied to existing LLL architectures to mitigate degradation. |
| Outcome: | The proposed method improves state-of-the-art models and reduces degradation compared to multi-task models. |
Meta-KD: A Meta Knowledge Distillation Framework for Language Model Compression across Domains (2021.acl-long)
Copied to clipboard
| Challenge: | Pre-trained language models have been successful in NLP tasks, but their large size and long inference time limit their deployment in real-time applications. |
| Approach: | They propose a meta-teacher model that captures transferable knowledge across domains and passes it to students. |
| Outcome: | The proposed model can distill large teacher models into small student models with guidance from the meta-teacher. |
Multilingual AMR Parsing with Noisy Knowledge Distillation (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Abstract Meaning Representation (AMR) parsing is a broad-coverage semantic formalism that encodes the meaning of a sentence as a rooted, directed, and labeled graph. |
| Approach: | They propose to use existing English parser to learn and improve multilingual AMR parsers . their results show that noisy input and precise output are key to successful distillation . |
| Outcome: | The proposed model outperforms the current state-of-the-art English-only parser on four different languages. |
GAML-BERT: Improving BERT Early Exiting by Gradient Aligned Mutual Learning (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to improve the early exiting of natural language processing (NLP) are notoriously gigantic and slow in both training and inference. |
| Approach: | They propose a framework for improving the early exiting of BERT by asking each exit to distill knowledge from each other. |
| Outcome: | The proposed framework outperforms the state-of-the-art (SOTA) BERT early exiting methods on the GLUE benchmark. |
Optimal Transport Guided Correlation Assignment for Multimodal Entity Linking (2024.findings-acl)
Copied to clipboard
Zefeng Zhang, Jiawei Sheng, Zhang Chuang, Liangyunzhi Liangyunzhi, Wenyuan Zhang, Siqi Wang, Tingwen Liu
| Challenge: | Existing methods to link ambiguous mentions to entities in multimodal knowledge graphs rely on partial correlations. |
| Approach: | They propose a framework that leverages multi-element correlations to bridge modality gap and enable fine-grained semantic matching by exploiting correlation between multimodal features and entities. |
| Outcome: | The proposed framework outperforms state-of-the-art models and confirms the effectiveness of the proposed method. |
DisCo: Distilled Student Models Co-training for Semi-supervised Text Mining (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing text mining models are fine-tuned by fine-timing a large pre-trained language model (PLM) in downstream tasks. |
| Approach: | They propose a semi-supervised learning framework for fine-tuning a cohort of small student models generated from a large pre-trained language model using knowledge distillation. |
| Outcome: | The proposed framework outperforms baseline models on semi-supervised text classification and extractive summarization tasks while maintaining comparable performance. |
Not to Overfit or Underfit the Source Domains? An Empirical Study of Domain Generalization in Question Answering (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to limit overfitting of training domains are rooted in this problem . domain generalization (DG) seeks to train models on a small number of source domains . |
| Approach: | They propose to use knowledge distillation to train models on a small number of source domains to maximize their zero-shot out-of-domain utility. |
| Outcome: | The proposed model learns its source domains better and has better out-of-domain generalization . the proposed model outperforms existing approaches that aim to limit overfitting . |
Improving Word Embedding Factorization for Compression Using Distilled Nonlinear Neural Decomposition (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Word-embeddings are vital components of natural language processing (NLP) but they consume a lot of memory which poses a challenge for edge deployment. |
| Approach: | They propose an embedding compression method based on matrix decomposition and knowledge distillation that initializes weights of pre-trained word-embeddings and fine-tunes end-to-end. |
| Outcome: | The proposed method has higher BLEU score on translation and lower perplexity on language modeling compared to complex, difficult to tune methods. |
HRKD: Hierarchical Relational Knowledge Distillation for Cross-domain Language Model Compression (2021.emnlp-main)
Copied to clipboard
| Challenge: | Large pre-trained language models (PLMs) have shown overwhelming performances on many tasks, but their large size and slow inference speed have hindered practical deployments. |
| Approach: | They propose a hierarchical relational knowledge distillation method to capture hierarchic and domain relational information. |
| Outcome: | The proposed method outperforms existing methods on multi-domain datasets and is highly reproducible. |
Combining Compressions for Multiplicative Size Scaling on Natural Language Tasks (2022.coling-1)
Copied to clipboard
| Challenge: | Quantization, knowledge distillation, and magnitude pruning are among the most popular methods for neural network compression in NLP. |
| Approach: | They compare accuracy vs. model size tradeoffs using quantization and distillation methods . they find that pruning provides greater benefit than quantization . |
| Outcome: | The proposed methods reduce model size and can accelerate inference, but their relative benefit and combinatorial interactions have not been rigorously studied. |
Can Small Language Models Help Large Language Models Reason Better?: LM-Guided Chain-of-Thought (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing frameworks for guiding a language model in reasoning tasks are limited by their tendency to generate low-quality rationales that are repetitive and vacuous. |
| Approach: | They propose a framework that leverages a lightweight language model for guiding a black-box large LM in reasoning tasks. |
| Outcome: | The proposed framework outperforms baselines in answer prediction accuracy. |
DiPair: Fast and Accurate Distillation for Trillion-Scale Text Matching and Pair Modeling (2020.findings-emnlp)
Copied to clipboard
Jiecao Chen, Liu Yang, Karthik Raman, Michael Bendersky, Jung-Jung Yeh, Yun Zhou, Marc Najork, Danyang Cai, Ehsan Emadzadeh
| Challenge: | Existing knowledge distillation models are not optimized for dealing with pairs (or tuples) of texts. |
| Approach: | They propose a framework for distilling fast and accurate models on text pair tasks using a scalable end-to-end training strategy. |
| Outcome: | Empirical studies on academic and real-world e-commerce benchmarks show the proposed framework can achieve speedups of over 350x and minimal quality drop relative to the cross-attention teacher BERT model. |
Ensembling and Knowledge Distilling of Large Sequence Taggers for Grammatical Error Correction (2022.acl-long)
Copied to clipboard
| Challenge: | Currently, machine translation (MT) is the mainstream approach for GEC. |
| Approach: | They propose to ensemble Transformer-based encoders by majority votes on span-level edits . their best ensemble achieves a new SOTA result even without pre-training on synthetic datasets - "Troy-Blogs" and "Try-1BW". |
| Outcome: | The proposed model achieves a new SOTA result even without pre-training on synthetic datasets. |
Aligned Weight Regularizers for Pruning Pretrained Neural Networks (2022.findings-acl)
Copied to clipboard
| Challenge: | Pruning aims to reduce the number of parameters while maintaining performance close to the original network. |
| Approach: | They propose a self-distilled pruning strategy that maximizes representational similarity between pruned and unpruned networks. |
| Outcome: | The proposed pruning strategy outperforms smaller models and outperformed smaller ones with an equal number of parameters and is competitive against (6 times) larger distilled networks. |
Consistent Representation Learning for Continual Relation Extraction (2022.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to train relation extraction models overfit memory samples and perform poorly on imbalanced datasets. |
| Approach: | They propose a method which uses contrastive learning and knowledge distillation to train a model on data with new relations while avoiding forgetting old ones. |
| Outcome: | The proposed method significantly outperforms state-of-the-art baselines and yields strong robustness on the imbalanced datasets. |
Knowledge Distillation ≈ Label Smoothing: Fact or Fallacy? (2023.emnlp-main)
Copied to clipboard
| Challenge: | Knowledge distillation (KD) is a method for knowledge transfer from one model to another . recent studies suggest it is based on label smoothing, but it is not . |
| Approach: | They propose to compare the predictive confidences of models trained with knowledge distillation . they propose to use a method that is similar to label smoothing to train models . |
| Outcome: | Experiments on four text classification tasks show that knowledge distillation and label smoothing drive model confidence in opposite directions. |
General Purpose Text Embeddings from Pre-trained Language Models for Scalable Inference (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Large pre-trained language models are currently used for many NLP tasks . however, inference for these models requires significant computational resources . |
| Approach: | They propose to use a shared text encoder to amortize the computational cost of inference over multiple tasks. |
| Outcome: | The proposed method reduces the size of the extracted representations by a factor of 16 to store them for later use. |
NeuroComparatives: Neuro-Symbolic Distillation of Comparative Knowledge (2024.findings-naacl)
Copied to clipboard
| Challenge: | Comparative knowledge is an essential component of our world knowledge, yet understudied in prior literature. |
| Approach: | They propose a framework for comparative knowledge distillation overgenerated from language models . they use a corpus of 8.8M comparisons over 1.74M entity pairs to acquire comparative information . |
| Outcome: | The proposed framework acquires comparative knowledge between everyday objects . human evaluations show that it outperforms existing resources in terms of validity . |
Efficient Transformer-based Large Scale Language Representations using Hardware-friendly Block Structured Pruning (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Pretrained large-scale language models have been criticized for their limited weight storage and computational speed on hardware platforms. |
| Approach: | They propose an efficient transformer-based large-scale language representation using hardware-friendly block structure pruning. |
| Outcome: | The proposed model achieves 5.0x accuracy on GLUE benchmarks and 1.79x compression rate on DistilBERT. |
LadaBERT: Lightweight Adaptation of BERT through Hybrid Model Compression (2020.coling-main)
Copied to clipboard
Yihuan Mao, Yujing Wang, Chufan Wu, Chen Zhang, Yang Wang, Quanlu Zhang, Yaming Yang, Yunhai Tong, Jing Bai
| Challenge: | Existing models that use knowledge distillation are memory-intensive and latency-prohibitive . Existing solutions that use this knowledge distilling framework are expensive . |
| Approach: | They propose a solution that uses weight pruning, matrix factorization and knowledge distillation to learn a smaller model. |
| Outcome: | The proposed model reduces the training overheads by an order of magnitude on public datasets while preserving state-of-the-art accuracy. |
Knowledge Inheritance for Pre-trained Language Models (2022.naacl-main)
Copied to clipboard
Yujia Qin, Yankai Lin, Jing Yi, Jiajie Zhang, Xu Han, Zhengyan Zhang, Yusheng Su, Zhiyuan Liu, Peng Li, Maosong Sun, Jie Zhou
| Challenge: | Existing large-scale pre-trained language models are mainly trained from scratch individually, ignoring that many well-taught PLMs are available. |
| Approach: | They propose a pre-training framework called knowledge inheritance and propose auxiliary supervision to efficiently learn larger PLMs. |
| Outcome: | The proposed framework can be used to train large-scale language models with huge parameters and a large dataset can be adapted to domain adaptation and knowledge transfer. |
Class-Incremental Few-Shot Event Detection (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing methods to deal with new class of events with only a few labeled instances are challenging . old knowledge forgetting and new class overfitting are two problems in this task. |
| Approach: | They propose a task called class-incremental few-shot event detection to solve old knowledge forgetting and new class overfitting problems. |
| Outcome: | The proposed method reduces old knowledge forgetting and new class overfitting problems on two benchmark datasets. |
Collective Wisdom: Improving Low-resource Neural Machine Translation using Adaptive Knowledge Distillation (2020.coling-main)
Copied to clipboard
| Challenge: | Existing approaches to train high-quality NMT models in bilingually low-resource scenarios are limited by the scarcity of parallel sentence-pairs. |
| Approach: | They propose to distill the knowledge of teacher models to a single student model by using knowledge distillation. |
| Outcome: | The proposed approach achieves up to +0.9 BLEU score improvements compared to strong baselines. |
MSD: Saliency-aware Knowledge Distillation for Multimodal Understanding (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Current knowledge distillation models are limited and lack performance on multimodal datasets. |
| Approach: | They propose a multimodal knowledge distillation framework to transfer knowledge from a teacher on multimodal tasks by learning the teacher's behavior within each modality. |
| Outcome: | The proposed framework achieves better performance than KD on four multimodal datasets. |
Collective Relevance Labeling for Passage Retrieval (2022.naacl-main)
Copied to clipboard
| Challenge: | Existing approaches to improve IR labels are incomplete and require computational overheads. |
| Approach: | They propose to distill knowledge for informed labeling without high computation overheads at evaluation time. |
| Outcome: | The proposed model outperforms state-of-the-art models while distilling the rankings better. |
Pruning-then-Expanding Model for Domain Adaptation of Neural Machine Translation (2021.naacl-main)
Copied to clipboard
| Challenge: | Existing methods for domain adaptation suffer from catastrophic forgetting, large domain divergence, and model explosion. |
| Approach: | They propose a method which prunes the model and keeps the important neurons or parameters responsible for both general-domain and in-domain translation. |
| Outcome: | The proposed method improves on different language pairs and domains compared with strong baselines. |
Multi-Task Learning with Shared Encoder for Non-Autoregressive Machine Translation (2021.naacl-main)
Copied to clipboard
| Challenge: | Existing non-autoregressive machine translation models have shown significant inference speedup but suffer from inferior translation accuracy. |
| Approach: | They propose to use AT as an auxiliary task to transfer AT knowledge to NAT models by knowledge distillation. |
| Outcome: | The proposed method achieves significant improvements over baseline non-Autoregressive machine translation models on WMT14 En-De and WMT16 En-Ro datasets. |
Distilling Knowledge for Empathy Detection (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Empathy is the link between self and others. |
| Approach: | They employ multi-task training with knowledge distillation to integrate knowledge from available resources to detect empathy from the natural language in different domains. |
| Outcome: | The proposed approach yields better results on an existing news-related empathy dataset compared to strong baselines. |
Static Word Embeddings for Sentence Semantic Representation (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods to learn fixed-length embeddings for sentence semantics require large computational cost, making it difficult to process billions of sentences cost-efficiently or deploy models on resource-constrained devices such as smartphones. |
| Approach: | They propose to extract word embeddings from a pre-trained Sentence Transformer and improve them with sentence-level principal component analysis followed by knowledge distillation or contrastive learning. |
| Outcome: | The proposed model outperforms existing models on sentence semantic tasks and surpasses a basic Sentence Transformer model (SimCSE) on a text embedding benchmark. |
Dual-teacher Knowledge Distillation for Low-frequency Word Translation (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Neural machine translation models are trained on parallel corpora with unbalanced word frequency distribution, resulting in high-frequency words being ignored. |
| Approach: | They propose to employ a low-frequency teacher model that excels in translating low- frequency words to guide the learning of the student model. |
| Outcome: | The proposed method achieves +0.64 BLEU improvements over the state-of-the-art method on the low-frequency translation task while maintaining the translation quality of high-frequency words. |
Mutual-Learning Improves End-to-End Speech Translation (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to end-to-end speech translation (E2E) models only allow one way knowledge transfer, which is limited by the performance of the teacher model. |
| Approach: | They propose a one-way knowledge transfer paradigm where the MT and ST models are collaboratively trained and considered as peers rather than teacher/student. |
| Outcome: | The proposed model improves the performance of end-to-end speech translation (ST) task by combining knowledge from two models with peer models. |
CoLaDa: A Collaborative Label Denoising Framework for Cross-lingual Named Entity Recognition (2023.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to named entity recognition (NER) are limited by the cost of labeling and labeling, especially for low-resource languages. |
| Approach: | They propose a model-collaboration-based denoising scheme that enables models trained on different data sources to collaboratively denoise pseudo labels used by each other. |
| Outcome: | The proposed framework achieves superior results on benchmark datasets and can generalize to distant languages. |
MTA4DPR: Multi-Teaching-Assistants Based Iterative Knowledge Distillation for Dense Passage Retrieval (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies have shown the effectiveness of knowledge distillation in DPR, but there is a performance gap between the teacher and the distilled student. |
| Approach: | They propose an iterative knowledge distillation method which transfers knowledge from teacher to student with help of multiple assistants in an iterated manner. |
| Outcome: | The proposed method achieves state-of-the-art performance among models with same parameters on multiple datasets and is competitive when compared with larger models. |
Scalable Syntax-Aware Language Models Using Knowledge Distillation (P19-1)
Copied to clipboard
| Challenge: | Prior work has shown that syntactic neural language models learn from small amounts of training data more effectively than sequential models. |
| Approach: | They propose a knowledge distillation technique that transfers knowledge from a syntactic language model trained on a small corpus to an LSTM language model and enables it to develop a more structurally sensitive representation of the larger training data. |
| Outcome: | The proposed method improves on baseline syntactic evaluations on LSTMs with a higher level of accuracy than previous methods. |
Claim Matching Beyond English to Scale Global Fact-Checking (2021.acl-long)
Copied to clipboard
| Challenge: | Existing methods to fact-check content are not scaled well in non-English contexts. |
| Approach: | They propose to use a WhatsApp tipline and public group message dataset to find pairs of textual messages containing claims that can be served with one fact-check. |
| Outcome: | The proposed model outperforms existing models in English, Hindi, and Tamil in all settings. |
Long-Tailed Question Answering in an Open World (2023.acl-long)
Copied to clipboard
| Challenge: | Existing QA approaches require access to seen tasks or do not explicitly model samples from unseen tasks. |
| Approach: | They propose an open-tailed QA model that encourages knowledge sharing between head, tail and unseen tasks and explicitly mines knowledge from a large pre-trained language model. |
| Outcome: | The proposed model outperforms the state-of-the-art on a large-scale dataset. |
ToXCL: A Unified Framework for Toxic Speech Detection and Explanation (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing models that focus on explicit toxic speech detection and explanation are prone to error propagation problems . et al., 2018) show that toxic speech models can be prone for generating errors . |
| Approach: | They propose a framework that can detect and explain toxic speech using a target group generator and an encoder-decoder model. |
| Outcome: | The proposed model outperforms baseline models and achieves state-of-the-art effectiveness . the proposed model generates a toxic explanation that matches the ground truth explanation . |
An Active Learning Framework for Inclusive Generation by Large Language Models (2025.coling-main)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) exhibit bias toward underrepresented groups, despite advances in active learning. |
| Approach: | They propose a clustering-based active learning framework enhanced with knowledge distillation that transforms the intermediate outputs of the learner model to yield more representative models without prior knowledge of underlying data distribution. |
| Outcome: | The proposed framework improves performance across data subgroups and lexical diversity, underscoring the model’s resilience to skewness in available data. |
PromptKD: Distilling Student-Friendly Knowledge for Generative Language Models via Prompt Tuning (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Recent advances in large language models (LLMs) have raised concerns about inference costs, increasing the need for research into model compression. |
| Approach: | They propose a method that utilizes prompt tuning to enable generative language models to transfer student-friendly knowledge. |
| Outcome: | Extensive experiments on instruction-following datasets show that PromptKD achieves state-of-the-art performance while adding only 0.0007% of the teacher’s parameters as prompts. |
Learn What Is Possible, Then Choose What Is Best: Disentangling One-To-Many Relations in Language Through Text-based Games (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Language models pre-trained on large self-supervised corpora, followed by task-specific fine-tuning has become the dominant paradigm in NLP. |
| Approach: | They propose to train language models pre-trained on large self-supervised corpora, followed by task-specific fine-tuning on the target domain. |
| Outcome: | The proposed model improves on the previous state-of-the-art model on the Jericho Walkthroughs dataset by 49%. |
BERM: Training the Balanced and Extractable Representation for Matching to Improve Generalization Ability of Dense Retrieval (2023.acl-long)
Copied to clipboard
| Challenge: | Dense retrieval has shown promise in the first-stage retrieval process when trained on in-domain labeled datasets. |
| Approach: | They propose a method to capture matching signal to improve generalization of dense retrieval by capturing matching signal between two texts. |
| Outcome: | The proposed method can be combined with different training methods to improve generalization ability without additional inference overhead and target domain data. |
DAdEE: Unsupervised Domain Adaptation in Early Exit PLMs (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Pre-trained Language Models (PLMs) exhibit good accuracy and generalization ability but their large size results in high inference latency. |
| Approach: | They propose an unsupervised domain adaptation framework that employs knowledge distillation to achieve domain-invariant representations at each layer. |
| Outcome: | The proposed framework outperforms early exit methods and domain adaptation methods under domain shift scenarios. |
LaCo: Large Language Model Pruning via Layer Collapse (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for model quantization, knowledge distillation, and model pruning are limited by hardware support limitations and the need for extensive training. |
| Approach: | They propose a layer-wise structured pruner that collapses rear model layers into a prior layer and enables a rapid reduction in model size while preserving the model structure. |
| Outcome: | The proposed pruner outperforms state-of-the-art pruning methods at pruning ratios of 25-30% and maintains an average task performance of over 80% at different pruning ratio. |
TinyBERT: Distilling BERT for Natural Language Understanding (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Pre-trained language models are computationally expensive and difficult to efficiently execute on resource-restricted devices. |
| Approach: | They propose a Transformer distillation method that performs Transformer distillations at pre-training and task-specific learning stages. |
| Outcome: | The proposed method accelerates inference and reduces model size while maintaining accuracy. |
CapEEN: Image Captioning with Early Exits and Knowledge Distillation (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Early Exit (EE) strategies can be used to enhance their efficiency, but their adaptation presents challenges in image captioning as it requires varying levels of semantic information for accurate predictions. |
| Approach: | They propose a framework to improve the performance of EE strategies by knowledge distillation . they use a variant A-CapEEN to adapt thresholds on the fly to account for drifts . |
| Outcome: | The proposed framework gains speedup of 1.77 while maintaining competitive performance compared to the final layer. |
How Does Distilled Data Complexity Impact the Quality and Confidence of Non-Autoregressive Machine Translation? (2021.findings-acl)
Copied to clipboard
| Challenge: | Prior work suggests that distilled training data is less complex than manual translations. |
| Approach: | They propose to use sequence-level knowledge distillation to match autoregressive models' translation quality. |
| Outcome: | The proposed model can match translation quality of autoregressive models with distilled training data. |
Distilling Calibrated Knowledge for Stance Detection (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for stance detection ignore meaningful signals among categories offered by hard labels. |
| Approach: | They propose to use knowledge distillation to calibrate teacher predictions in each generation step. |
| Outcome: | The proposed method can calibrate teacher predictions in each generation step and improves stance detection accuracy. |
Learning from Imperfect Data: Towards Efficient Knowledge Distillation of Autoregressive Language Models for Text-to-SQL (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing text-to-SQL LLMs are computationally expensive and difficult to deploy in real-world applications. |
| Approach: | They propose to distill a larger teacher model into a smaller student model by using imperfect data to improve the KD. |
| Outcome: | The proposed method achieves the best tradeoff between performance and efficiency on 5 text-to-SQL benchmarks. |
Data Laundering: Artificially Boosting Benchmark Results through Knowledge Distillation (2025.acl-long)
Copied to clipboard
| Challenge: | Existing studies show that language model benchmarks are vulnerable to manipulation and exploitation. |
| Approach: | They propose a method that allows the covert transfer of benchmark-specific knowledge through seemingly legitimate intermediate training steps. |
| Outcome: | The proposed method can achieve significant improvements in accuracy without developing reasoning capabilities. |
Evaluating the Expressive Appropriateness of Speech in Rich Contexts (2026.acl-long)
Copied to clipboard
Tianrui Wang, Ziyang Ma, Yizhou Peng, Haoyu Wang, Zhikang Niu, Zikang Huang, Yihao Wu, Yi-Wen Chao, Yu Jiang, Yuheng Lu, Guanrou Yang, Xuanchen Li, Hexin Liu, Chunyu Qiang, Cheng Gong, Yifan Yang, Tianchi Liu, Junyu Wang, Nana Hou, Meng Ge, Fuming You, Yang Wei, Zhongqian Sun, Hu Haifeng, Xiaobao Wang, Eng Siong Chng, Xie Chen, Longbiao Wang, Jianwu Dang
| Challenge: | Existing methods for evaluating expressive speech focus on word accuracy, naturalness, signal quality, or emotional intensity at the utterance level. |
| Approach: | They propose a framework for Evaluating Expressive Appropriateness in speech that assesses whether a speech sample aligns with the underlying communicative intent implied by its discourse-level narrative context. |
| Outcome: | The proposed framework outperforms existing speech evaluation and analysis systems on a human-annotated test set. |
SKD-NER: Continual Named Entity Recognition via Span-based Knowledge Distillation with Reinforcement Learning (2023.emnlp-main)
Copied to clipboard
| Challenge: | Continual learning for named entity recognition (CL-NER) aims to enable models to continuously learn new entity types while retaining the ability to recognize previously learned ones. |
| Approach: | They propose a model that leverages knowledge distillation to retain memory and employs reinforcement learning strategies to optimize the soft labeling and distillation losses generated by the teacher model to effectively prevent catastrophic forgetting. |
| Outcome: | The proposed model outperforms state-of-the-art models on two benchmark datasets showing that it significantly improves the performance of the CL-NER task. |
CILDA: Contrastive Data Augmentation Using Intermediate Layer Knowledge Distillation (2022.coling-1)
Copied to clipboard
Md Akmal Haidar, Mehdi Rezagholizadeh, Abbas Ghaddar, Khalil Bibi, Phillippe Langlais, Pascal Poupart
| Challenge: | Knowledge distillation (KD) is an efficient framework for compressing large-scale pre-trained language models. |
| Approach: | They propose a data augmentation technique tailored for knowledge distillation based on contrastive loss to improve masked adversarial data augmented by intermediate layer matching. |
| Outcome: | The proposed technique outperforms state-of-the-art methods on the GLUE benchmark and in an out-of domain evaluation. |
Pro-KD: Progressive Distillation by Following the Footsteps of the Teacher (2022.coling-1)
Copied to clipboard
| Challenge: | Knowledge distillation (KD) is a powerful tool for deep learning applications. |
| Approach: | They propose a method which defines a smoother training path for the student by following the training footprints of the teacher rather than solely relying on distilling from a single mature fully-trained teacher. |
| Outcome: | The proposed technique is quite effective in mitigating the capacity-gap problem and the checkpoint search problem. |
Attend, Select and Eliminate: Accelerating Multi-turn Response Selection with Dual-attention-based Content Elimination (2023.findings-acl)
Copied to clipboard
| Challenge: | Pre-trained language models can be used to perform multi-turn response selection, but they can be expensive. |
| Approach: | They propose a framework and a strategy that progressively selects and eliminates unimportant content under context-response dual-attention. |
| Outcome: | The proposed method can effectively speed-up SOTA models without much performance degradation and shows a better trade-off between speed and performance than previous methods. |
Iterative Structured Knowledge Distillation: Optimizing Language Models Through Layer-by-Layer Distillation (2025.coling-main)
Copied to clipboard
| Challenge: | Structured pruning and knowledge distillation are often not efficient and require a fixed architecture, limiting flexibility. |
| Approach: | They propose a method which integrates knowledge distillation and structured pruning by replacing transformer blocks with smaller, efficient versions during training. |
| Outcome: | The proposed method outperforms L1 pruning and maintains four-fifths of performance on language modeling and commonsense reasoning tasks. |
LLM Distillation for Efficient Few-Shot Multiple Choice Question Answering (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large Language Models excel at few-shot learning but their direct application in real-world scenarios is often hindered by their high computational cost. |
| Approach: | They propose a framework that uses Large Language Models for data generation and scoring to improve encoder model performance. |
| Outcome: | The proposed approach improves accuracy from 28.9% to 39.3% on a few-shot MCQA task . |
Multi-Grained Knowledge Distillation for Named Entity Recognition (2021.naacl-main)
Copied to clipboard
| Challenge: | Pre-trained big models have delivered top performance in Seq2seq modeling, but their deployments in real-world applications are often hindered by excessive computations and memory demands. |
| Approach: | They propose a distillation scheme to efficiently transfer knowledge from big models to their cheaper counterparts. |
| Outcome: | The proposed scheme maximizes the assimilation of knowledge from the teacher model to the student model. |
Learning to Plan for Retrieval-Augmented Large Language Models from Knowledge Graphs (2024.findings-emnlp)
Copied to clipboard
Junjie Wang, Mingyang Chen, Binbin Hu, Dan Yang, Ziqi Liu, Yue Shen, Peng Wei, Zhiqiang Zhang, Jinjie Gu, Jun Zhou, Jeff Pan, Wen Zhang, Huajun Chen
| Challenge: | Recent studies have attempted to enhance the performance of large language models (LLMs) in complex question-answering (QA) tasks by combining step-wise planning with external retrieval. |
| Approach: | They propose a framework for enhancing LLMs’ planning capabilities by using planning data derived from knowledge graphs (KGs). |
| Outcome: | The proposed framework improves LLMs’ planning capabilities by using knowledge graphs (KGs) the proposed framework is compared with existing frameworks on multiple datasets and shows that it is effective for large language models. |
Improved Knowledge Distillation for Pre-trained Language Models via Knowledge Selection (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing studies on knowledge distillation have shown that not all knowledge is necessary for learning a good student model. |
| Approach: | They propose an actor-critic approach to selecting appropriate knowledge to transfer during the process of knowledge distillation. |
| Outcome: | The proposed method outperforms several strong knowledge distillation baselines significantly on the GLUE datasets. |
Task-Attentive Transformer Architecture for Continual Learning of Vision-and-Language Tasks Using Knowledge Distillation (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing algorithms for learning unimodal vision-only or language-only tasks are limited by the size and computational load of fine-tuning large-scale pre-trained neural networks. |
| Approach: | They propose a transformer-based CL architecture for learning bimodal vision-and-language tasks by increasing the number of the learnable parameters dynamically and using knowledge distillation. |
| Outcome: | The proposed model reaches state-of-the-art on vision-and-language tasks. |
POS-Constrained Parallel Decoding for Non-autoregressive Generation (2021.acl-long)
Copied to clipboard
| Challenge: | Existing non-autoregressive generation systems face multimodality problem due to conditionally independent decoding. |
| Approach: | They propose to incorporate linguistic structure into NAG inference instead of teacher AG . they propose a method that provides a specific POS sequence to constrain the NAG model . |
| Outcome: | The proposed method improves NAG models on four text generation tasks to a greater extent compared to knowledge distillation. |
Assist Non-native Viewers: Multimodal Cross-Lingual Summarization for How2 Videos (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing multimodal summarization methods are limited to monolingual videos . a proposed task aims to generate cross-lingual summaries from multimodal inputs . |
| Approach: | They propose a task to generate cross-lingual summaries from multimodal inputs of videos . they propose fusion network that integrates multimodal and cross-linguistic information . |
| Outcome: | The proposed task outperforms existing methods on a reorganized How2 dataset on the reorganized How2 data set. |
ConGen: Unsupervised Control and Generalization Distillation For Sentence Representation (2022.findings-emnlp)
Copied to clipboard
Peerat Limkonchotiwat, Wuttikorn Ponwitayarat, Lalita Lowphansirikul, Can Udomcharoenchaikit, Ekapol Chuangsuwanich, Sarana Nutanong
| Challenge: | Sentence representations are essential in many NLP tasks operating at the sentence level. |
| Approach: | They propose an unsupervised sentence representation method to reduce the supervised-unsupervised performance gap for smaller models. |
| Outcome: | The proposed method outperforms supervised training on STS, text classification, and natural language inference tasks on smaller models. |
BERT Learns to Teach: Knowledge Distillation with Meta Learning (2022.acl-long)
Copied to clipboard
| Challenge: | Existing knowledge distillation methods are based on teacher model, but have drawbacks . a teacher model is fixed during training, but meta learning can improve student performance . |
| Approach: | They propose a meta learning framework that allows the teacher network to learn to better transfer knowledge to the student network. |
| Outcome: | Experiments show that MetaDistil can improve on existing methods and is less sensitive to student capacity and hyperparameters. |
Knowledge Base Embedding By Cooperative Knowledge Distillation (2020.coling-main)
Copied to clipboard
| Challenge: | Knowledge bases are increasingly exploited as gold standard data sources for various knowledge-driven NLP tasks. |
| Approach: | They propose a method to perform knowledge base representation learning by mutually and jointly distilling knowledge within a dynamic teacher-student setting. |
| Outcome: | The proposed approach outperforms two baselines, traditional and sequential, on two standard datasets showing that it is possible to distill knowledge between KBs. |
SMASH: Improving SMAll Language Models’ Few-SHot Ability with Prompt-Based Distillation (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Large-scale language models with prompts have shown remarkable performance on few-shot learning. |
| Approach: | They propose an approach to improve SMAll language models’ few-SHot ability by training on intermediate tasks before prompt-based fine-tuning on downstream tasks. |
| Outcome: | The proposed model improves on sentence-pair and sentiment classification tasks by training on intermediate tasks before fine-tuning on downstream tasks. |
Autoregressive Knowledge Distillation through Imitation Learning (2020.emnlp-main)
Copied to clipboard
| Challenge: | Autoregressive models are ubiquitous in natural language processing due to the sequential nature of text generation. |
| Approach: | They propose a compression technique for autoregressive models driven by an imitation learning perspective on knowledge distillation. |
| Outcome: | The proposed method outperforms other distillation algorithms on translation and summarization tasks while increasing inference speed 14 times. |
Tutoring Helps Students Learn Better: Improving Knowledge Distillation for BERT with Tutor Network (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing knowledge distillation approaches for language models have overlooked the difficulty of training examples. |
| Approach: | They propose a framework that controls difficulty of training examples during pre-training by a tutor network. |
| Outcome: | The proposed framework outperforms state-of-the-art KD methods with student models on the GLUE benchmark. |
Selective Knowledge Distillation for Neural Machine Translation (2021.acl-long)
Copied to clipboard
| Challenge: | Neural Machine Translation models achieve state-of-the-art performance on many translation benchmarks. |
| Approach: | They propose a protocol that analyzes different impacts of samples by comparing various samples’ partitions. |
| Outcome: | The proposed methods yield up to +1.28 and +0.89 BLEU points improvements over the Transformer baseline, respectively. |
Improving Stance Detection with Multi-Dataset Learning and Knowledge Distillation (2021.emnlp-main)
Copied to clipboard
| Challenge: | stance detection is a method to determine whether a text author is in favor of, against or neutral toward a specific target. |
| Approach: | They propose a method that applies instance-specific temperature scaling to the teacher and student predictions. |
| Outcome: | The proposed method outperforms the state-of-the-art on all datasets and on multiple datasets. |
Protecting Language Models Against Unauthorized Distillation through Trace Rewriting (2026.acl-long)
Copied to clipboard
| Challenge: | Knowledge distillation is a widely adopted technique for transferring capabilities from LLMs to smaller, more efficient student models. |
| Approach: | They propose methods for modifying teacher-generated reasoning traces to achieve two objectives that deter unauthorized distillation: (1) anti-distillation, or degrading the training usefulness of query responses; and (2) API watermarking, which embeds verifiable signatures in student models. |
| Outcome: | The proposed method achieves strong anti-distillation effect while maintaining or even improving teacher performance. |
Scalable Zero-shot Entity Linking with Dense Entity Retrieval (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for entity linking use manually curated mention tables and incoming Wikipedia link popularity. |
| Approach: | They propose a BERT-based entity linking model with a bi-encoder that embeds the mention context and the entity descriptions and then re-ranked the candidate with . they also evaluate the accuracy-speed trade-off inherent to large pre-trained models. |
| Outcome: | The proposed model is state-of-the-art on recent zero-shot benchmarks and established non-zero-shot evaluations. |
Enhancing Systematic Decompositional Natural Language Inference Using Informal Logic (2024.emnlp-main)
Copied to clipboard
Nathaniel Weir, Kate Sanders, Orion Weller, Shreya Sharma, Dongwei Jiang, Zhengping Jiang, Bhavana Dalvi Mishra, Oyvind Tafjord, Peter Jansen, Peter Clark, Benjamin Van Durme
| Challenge: | Recent language models allow structured reasoning with text, but lack of a clear protocol for discerning entailment causes noisy datasets and limited performance gains. |
| Approach: | They propose a consistent approach to annotating decompositional entailment and evaluate its impact on LLM-based textual inference. |
| Outcome: | The proposed approach has higher internal consistency than prior decompositional entailment datasets and significantly improves proof quality and accuracy. |
Bridging Fairness and Environmental Sustainability in Natural Language Processing (2022.emnlp-main)
Copied to clipboard
| Challenge: | a lack of research on the interplay between fairness and environmental impact is a problem in natural language processing . fairness is prone to encode and amplify stereotypical social biases, according to several studies . |
| Approach: | They evaluate a technique to reduce energy consumption of English NLP models by knowledge distillation for its impact on fairness. |
| Outcome: | The proposed method reduces energy consumption and environmental impact of English NLP models. |
CTC-based Non-autoregressive Textless Speech-to-Speech Translation (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing direct speech-to-speech translation models require text supervision during training, which is not feasible for numerous unwritten languages. |
| Approach: | They propose a non-autoregressive (NAR) model that generates discrete units from the source speech and employs a unit-based vocoder to synthesize the target. |
| Outcome: | The proposed model achieves translation quality comparable to the autoregressive model while preserving up to 26.81 decoding speedup. |
DistillCSE: Distilled Contrastive Learning for Sentence Embeddings (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to sentence embeddings are based on contrastive learning (CL) . |
| Approach: | They propose a framework which performs contrastive learning under the self-training paradigm with knowledge distillation and propose 'Group-P shuffling strategy' and averaging logits from multiple teacher components. |
| Outcome: | The proposed framework outperforms many strong baseline methods and yields a new state-of-the-art performance. |
Momentum Posterior Regularization for Multi-hop Dense Retrieval (2025.coling-main)
Copied to clipboard
| Challenge: | Current methods for knowledge distillation in one-time retrieval are ineffective for multi-hop QA . posterior information is often defined as the response, which may not connect to the query without intermediate retrieval . |
| Approach: | They propose to distill knowledge from a posterior retrieval into a prior retrieval for multi-hop QA . they propose to use momentum moving average method to update posterior information along with prior retrievals . |
| Outcome: | Experiments on HotpotQA and StrategyQA show that MoPo outperforms baselines in retrieval and downstream QA tasks. |
CLASSIC: Continual and Contrastive Learning of Aspect Sentiment Classification Tasks (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies have focused on continual learning of aspect sentiment classification (ASC) tasks in domain incremental learning (DIL) |
| Approach: | They propose a continual learning method that learns a sequence of tasks incrementally . they propose CLASSIC, which uses a domain incremental learning setting . |
| Outcome: | The proposed model is highly effective in a domain incremental learning setting. |
Distilling Structured Knowledge for Text-Based Relational Reasoning (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing text-based relational reasoning models lack a symbolic representation of text . performance gap between NLP models and structured models remains . |
| Approach: | They first pre-train a GNN on a reasoning task using structured inputs and then incorporate its knowledge into an NLP model. |
| Outcome: | The proposed model improves on two state-of-the-art NLP models on 13 different inductive reasoning datasets from the CLUTRR benchmark. |
Students Who Study Together Learn Better: On the Importance of Collective Knowledge Distillation for Domain Transfer in Fact Verification (2021.emnlp-main)
Copied to clipboard
| Challenge: | Neural networks depend heavily on lexicalized information, which can be overfitted . this can be a problem in fact verification, which has important societal implications. |
| Approach: | They propose a knowledge distillation approach for fact verification using student models. |
| Outcome: | The proposed approach outperforms state-of-the-art classifiers on a training dataset and in supervised settings. |
Leveraging Entity Information for Cross-Modality Correlation Learning: The Entity-Guided Multimodal Summarization (2024.findings-acl)
Copied to clipboard
| Challenge: | Multimodal Summarization with Multimodal Output (MSMO) is a new approach to produce a multimodal summary that integrates both text and relevant images. |
| Approach: | They propose an Entity-Guided Multimodal Summarization model that integrates both text and relevant images to produce a multimodal summary. |
| Outcome: | The proposed model integrates text-image and entity-image information and refines image selection through knowledge distillation from a pre-trained vision-language model. |
Distilling the Essence: Efficient Reasoning Distillation via Sequence Truncation (2026.findings-acl)
Copied to clipboard
Wei-Rui Chen, Vignesh Kothapalli, Ata Fatahibaarzi, Hejian Sang, Shao Tang, Qingquan Song, Zhipeng Wang, Muhammad Abdul-Mageed
| Challenge: | a large reasoning model (LRM) training on large amounts of reasoning data is computationally expensive. |
| Approach: | They propose a method to quantify computation-quality tradeoffs as a function of sequence length. |
| Outcome: | The proposed method reduces training time, memory and FLOPs by 50% on long training sequences while retaining the full-sequence performance. |
Continual Learning with Semi-supervised Contrastive Distillation for Incremental Neural Machine Translation (2024.acl-long)
Copied to clipboard
| Challenge: | Multi-domain learning is a good solution for solving domain tasks but it requires retraining when adding a new domain. |
| Approach: | They propose to exploit unlabeled data from the same distributions of the older domains to avoid catastrophic forgetting. |
| Outcome: | The proposed framework exploits unlabeled data from the same distributions of the older domains to avoid catastrophic forgetting. |
BAM! Born-Again Multi-Task Networks for Natural Language Understanding (P19-1)
Copied to clipboard
| Challenge: | Existing methods to train multi-task neural networks outperform or even match their single-task counterparts are difficult to implement. |
| Approach: | They propose a method that uses knowledge distillation to train multi-task neural networks that outperform or even match their single-task counterparts. |
| Outcome: | The proposed method outperforms or matches single-task neural networks on the GLUE benchmark. |
Universal-KD: Attention-based Output-Grounded Intermediate Layer Knowledge Distillation (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for intermediate layer matching are limited due to huge over-parameterization . |
| Approach: | They propose to match intermediate layers of teacher and student in output space via attention-based layer projection. |
| Outcome: | The proposed method outperforms existing methods on GLUE tasks. |
Domain Adaptation for Conversational Query Production with the RAG Model Feedback (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing studies have focused on human-annotated search queries but they can not cover conversations of various domains. |
| Approach: | They propose a domain adaptation framework that uses retrieval-augmented generation to improve the model's robustness. |
| Outcome: | The proposed model is more robust and performs significantly better in a more challenging setting over strong baselines. |
Are Intermediate Layers and Labels Really Necessary? A General Language Model Distillation Method (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing knowledge distillation methods rely on intermediate layer features and golden labels, which require aligned model architecture and labeled data respectively. |
| Approach: | They propose a general language model distillation method that performs two-stage word prediction distillation and vocabulary compression, which is simple and shows extremely strong performance. |
| Outcome: | The proposed method outperforms 25 state-of-the-art methods on the SuperGLUE benchmark, achieving an average score that surpasses the best method by 3%. |
Language Model Prior for Low-Resource Neural Machine Translation (2020.emnlp-main)
Copied to clipboard
| Challenge: | Neural machine translation is based on large parallel corpora and requires expensive training and training. |
| Approach: | They propose to incorporate a LM as prior in a neural translation model (TM) they add a regularization term which pushes the output distributions to be probable under the LM prior . |
| Outcome: | The proposed approach does not compromise decoding speed, because the LM is used only at training time, unlike previous work that requires it during inference. |
Calibrating Student Models for Emotion-related Tasks (2022.emnlp-main)
Copied to clipboard
| Challenge: | Knowledge distillation is an effective method to transfer knowledge from one network (a.k.a. teacher) to another (as student). |
| Approach: | They propose to use a mixup data augmentation technique to increase the accuracy of the model by providing better training signals to the student models. |
| Outcome: | The proposed method improves the calibration of student models while providing better training signals to the student models using training dynamics. |
RUIE: Retrieval-based Unified Information Extraction using Large Language Model (2025.coling-main)
Copied to clipboard
| Challenge: | Unified information extraction (UIE) aims to extract diverse structured information from unstructured text using a single model or framework. |
| Approach: | They propose a framework that leverages in-context learning for efficient task generalization by combining LLM preferences with a keyword-enhanced reward model. |
| Outcome: | The proposed framework performs better on eight held-out datasets than existing methods and instruction-tuning methods. |
Can LLM Watermarks Robustly Prevent Unauthorized Knowledge Distillation? (2025.acl-long)
Copied to clipboard
| Challenge: | Large Language Model (LLM) watermarking is radioactive and enables the detection of watermarks inherited by student models when trained on the outputs of watermarked teacher models. |
| Approach: | They propose two types of watermark removal attacks that allow student models to perform untraceable knowledge distillation while avoiding watermark inheritance. |
| Outcome: | The proposed attacks eliminate inherited watermarks while maintaining knowledge transfer efficiency and low computational overhead. |
Adaptive Label Smoothing with Self-Knowledge in Natural Language Generation (2022.emnlp-main)
Copied to clipboard
| Challenge: | Overconfidence in model generalization and calibration has been shown to impair model generalisation and calibration. |
| Approach: | They propose a regularization scheme that takes model probability into account and takes it into account . they use a prior label distribution to smooth target labels . |
| Outcome: | The proposed model improves model generalization and calibration by taking model probability into account. |
Modular Transformers: Compressing Transformers into Modularized Layers for Flexible Efficient Inference (2023.findings-acl)
Copied to clipboard
| Challenge: | Pre-trained sequence-to-sequence models have advanced the state of the art on text generation tasks. |
| Approach: | They introduce a modular encoder-decoder framework for flexible sequence-to-sequence model compression. |
| Outcome: | The proposed framework can achieve flexible compression ratios from 1.1x to 6x with little to moderate relative performance drop. |
Hard Gate Knowledge Distillation - Leverage Calibration for Robust and Reliable Language Model (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing knowledge distillation schemes focus on a teacher as a source of knowledge and a gauge to detect miscalibration of a student. |
| Approach: | They propose a method that uses a teacher model as a source of knowledge and a model as an error detector to detect miscalibration of a student. |
| Outcome: | The proposed scheme improves model generalization and significantly lowers calibration error. |
ADAM: Dense Retrieval Distillation with Adaptive Dark Examples (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to retrieve data from multiple encoders are too trivial for the teacher to distinguish, preventing the teacher from transferring abundant dark knowledge to the student. |
| Approach: | They propose a knowledge distillation framework that can better transfer the dark knowledge held in the teacher with adaptive dark examples. |
| Outcome: | The proposed framework can better transfer the dark knowledge held in the teacher with adaptive dark examples. |
A Study on Knowledge Distillation from Weak Teacher for Scaling Up Pre-trained Language Models (2023.findings-acl)
Copied to clipboard
| Challenge: | a study shows that DWT can be effective in the vision domain and natural language processing pre-training stages. |
| Approach: | They examine three key factors to optimize Distillation from Weak Teacher (DWT) DWT is a method of transferring knowledge from a weaker teacher model to a larger student model to improve its performance. |
| Outcome: | a new study examines three key factors to optimize DWT in NLP pre-training scenarios . the impact of teacher model quality and guidelines for adjusting the weighting value for DW T loss are examined . |
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. |
3DM: Distill, Dynamic Drop, and Merge for Debiasing Multi-modal Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent advances in Multi-modal Language Models have shown remarkable performance in multimodal tasks . however, these models often exhibit inherent biases that compromise their reliability and fairness. |
| Approach: | They propose a framework that integrates Distill, Dynamic Drop, and Merge to address these challenges. |
| Outcome: | The proposed framework outperforms existing methods in balancing debiasing and improving performance on the MMSD2.0 sarcasm detection dataset. |
Exploring Non-Autoregressive Text Style Transfer (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for text style transfer use autoregressive decoding, but they are slow and low parallelizability. |
| Approach: | They propose a base NAR model by directly adapting the common training scheme from its AutoRegressive counterpart. |
| Outcome: | The proposed model sacrifices performance due to lack of conditional dependence between output tokens . knowledge distillation, contrastive learning, and iterative decoding are employed to improve the model . |
Bridging the Gap between Decision and Logits in Decision-based Knowledge Distillation for Pre-trained Language Models (2023.acl-long)
Copied to clipboard
| Challenge: | Existing knowledge distillation methods require access to internal information of teachers . however, such information is not always accessible for large pre-trained language models . |
| Approach: | They propose a method to estimate logits from the decision distributions using logits theoretically and empirically. |
| Outcome: | The proposed method outperforms baselines on natural language understanding and machine reading comprehension datasets. |
Logits-Based Finetuning (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for developing compact and efficient large language models lack token-level dependencies and linguistic diversity. |
| Approach: | They propose a logits-based fine-tuning framework that integrates supervised learning and knowledge distillation to build enriched training targets using teacher logits and ground truth labels. |
| Outcome: | The proposed method outperforms existing methods on a large-scale logits dataset and a series of science-focused models. |
Select, Prompt, Filter: Distilling Large Language Models for Summarizing Conversations (2023.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) can be expensive to train, deploy, and use for specific natural language generation tasks. |
| Approach: | They propose a method to distill ChatGPT and fine-tune smaller LMs for summarizing forum conversations using a semantic similarity metric. |
| Outcome: | The proposed method leads to significant improvements of up to 6.6 ROUGE-2 score by leveraging sufficient in-domain pseudo-labeled data over standard KD approach given the same size of training data. |
Temperature-Centric Investigation of Speculative Decoding with Knowledge Distillation (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Speculative decoding is a novel method to expedite inference in autoregressive (large) language models. |
| Approach: | They propose to use a smaller model as a draft model to speculate a block of tokens, which the target model then evaluates for acceptance. |
| Outcome: | The proposed method can be used to accelerate inference in autoregressive (large) language models by using smaller models as draft models to speculate tokens for multiple inference steps. |
Evaluating distillation methods for data-efficient syntax learning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | knowledge distillation (KD) targeting attention should selectively accelerate syntax acquisition, a study finds . logit-based KD dramatically improves data-efficiency, attention-based one provides minimal benefit even for syntactic tasks. |
| Approach: | a study predicts that knowledge distillation targeting attention should selectively accelerate syntax acquisition . a systolic analysis of student models compared to logit-based knowledge distillations . |
| Outcome: | a new study shows that knowledge distillation (KD) targeting attention accelerates syntax acquisition . the hypothesis is tested on syntactic benchmarks and perplexity. |
Better Language Models of Code through Self-Improvement (2023.findings-acl)
Copied to clipboard
| Challenge: | Pre-trained language models for code (PLMCs) are pre-taught on large datasets with multi-modal objectives, but fine-tuning them requires extensive supervision and is limited by the size of the dataset provided. |
| Approach: | They propose a data augmentation framework that utilizes knowledge from the pre-training and fine-tuning stage to augment training data, which is then used for the next step. |
| Outcome: | The proposed framework significantly improves pre-trained language models’ performance in sequence-generation tasks, such as code summarization and code generation in the CodeXGLUE benchmark. |
Beyond Preserved Accuracy: Evaluating Loyalty and Robustness of BERT Compression (2021.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies on compression of pretrained language models usually use preserved accuracy as the metric for evaluation. |
| Approach: | They propose two new metrics that measure how closely a compressed model mimics the original model. |
| Outcome: | The proposed metrics measure how closely a compressed model (i.e., student) mimics the original model (e.g., teacher). |
PruMUX: Augmenting Data Multiplexing with Model Compression (2023.findings-acl)
Copied to clipboard
| Challenge: | Prior work has investigated methods like model pruning, knowledge distillation, and data multiplexing to increase model throughput without sacrificing accuracy. |
| Approach: | They propose to combine structured pruning and data multiplexing methods to increase model throughput without sacrificing accuracy. |
| Outcome: | The proposed method achieves 7.5-29.5X throughput improvement over a BERT-base model with accuracy threshold from 80% to 74%. |
EfficientVLM: Fast and Accurate Vision-Language Models via Knowledge Distillation and Modal-adaptive Pruning (2023.findings-acl)
Copied to clipboard
| Challenge: | Pre-trained vision-language models have achieved impressive results in a range of vision-linguistic tasks. |
| Approach: | They propose a distilling then pruning framework to compress large vision-language models into smaller, faster ones. |
| Outcome: | The proposed framework reduces the size of a pre-trained large vision-language model and improves its performance on vision-linguistic tasks. |
Beyond the Tip of Efficiency: Uncovering the Submerged Threats of Jailbreak Attacks in Small Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Small language models (SLMs) have become increasingly prominent in the deployment on edge devices due to their high efficiency and low computational cost. |
| Approach: | They evaluate the security performance of 13 state-of-the-art small language models under various jailbreak attacks. |
| Outcome: | The proposed methods demonstrate that SLMs are quite susceptible to jailbreak attacks and some are even vulnerable to harmful prompts. |
Consistency is Key: On Data-Efficient Modality Transfer in Speech Translation (2023.findings-emnlp)
Copied to clipboard
| Challenge: | End-to-end approaches to speech translation suffer from data scarcity compared to machine translation (MT). |
| Approach: | They propose a method which combines knowledge distillation and consistency learning to break the dilemma of learning-forgetting. |
| Outcome: | The proposed method outperforms the previous methods on a MuST-C dataset even without additional data. |
LLMR: Knowledge Distillation with a Large Language Model-Induced Reward (2024.lrec-main)
Copied to clipboard
| Challenge: | Large language models have demonstrated remarkable performance in various NLP tasks, but are typically computationally expensive and difficult to be deployed in resource-constrained environments. |
| Approach: | They propose a knowledge distillation method based on a reward function induced from large language models. |
| Outcome: | The proposed method outperforms traditional methods on multiple datasets and tasks. |
Failures are Treasures: Constructing a Pedagogical Bridge for Agentic Strategy Distillation (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing knowledge distillation methods focus on imitating successful trajectories, whereas small language models are fragile and often collapsing after encountering errors. |
| Approach: | They propose a Pedagogical Bridge for Reflective Insight and Distillation of Guiding Errors that combines reflection-in-action and reflection-on-action to enable agents to diagnose and correct critical errors while abstracting transferable strategies from contrastive student–teacher trajectories. |
| Outcome: | Experiments show that the proposed model significantly elevates performance in large language models (SLMs) . |
DIMSIM: Distilled Multilingual Critics for Indic Text Simplification (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to improve the quality of responses generated by large language models (LLMs) however, these critique-refine steps require multiple expensive LLM calls. |
| Approach: | They propose to use critique distillation to train critic models that are trained on input-critique pairs generated by an LLM. |
| Outcome: | The proposed model trains two separate critics that focus on lexical and structure complexity, and is more effective than using an LLM directly as a critic in both 0-shot and few-shot settings. |
MinT: Boosting Generalization in Mathematical Reasoning via Multi-view Fine-tuning (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing methods focus on specializing LMs in mathematical reasoning and rely on knowledge distillation. |
| Approach: | They propose a multi-view fine-tuning method that exploits existing mathematical problem datasets with diverse annotation styles. |
| Outcome: | The proposed method outperforms existing methods that rely heavily on LLM teachers . it grants models generalization ability across views and datasets, and the capability to learn from inaccurate or incomplete data. |
D2TV: Dual Knowledge Distillation and Target-oriented Vision Modeling for Many-to-Many Multimodal Summarization (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing studies focus on improving MMS models by filtering summary-unrelated visual features with implicit learning or explicitly complex training objectives. |
| Approach: | They propose a multimodal multimodal summarization task that aims to generate summaries in any language with document inputs in any languages and the corresponding image sequence. |
| Outcome: | The proposed task can generate summaries in any language with document inputs in any languages and the corresponding image sequence. |
SelFusion: Self-distillation for Diffusion Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Existing knowledge distillation methods for autoregressive large language models (LLMs) are not effective for reducing generation quality, but they can be useful for real-time applications. |
| Approach: | They propose a self-distillation framework that allows for effective KD without external teacher . they propose to use two modes of knowledge distillation to determine distillation direction . |
| Outcome: | The proposed framework outperforms existing methods with external teachers on instruction-following tasks. |
Model Calibration for Emotion Detection (2025.findings-emnlp)
Copied to clipboard
| Challenge: | a MixUp method is used to calibrate emotion detection models based on knowledge distillation and the MixUp data augmentation technique. |
| Approach: | They propose a method that uses knowledge distillation and the MixUp data augmentation technique to enhance the trustworthiness of emotion detection models. |
| Outcome: | The proposed method improves the accuracy of the teacher models and the student models. |
PIRB: A Comprehensive Benchmark of Polish Dense and Hybrid Text Retrieval Methods (2024.lrec-main)
Copied to clipboard
| Challenge: | PIRB is a framework for text information retrieval in Polish . existing and new datasets are evaluated to evaluate the performance of 41 models . |
| Approach: | They propose a framework for 41 text information retrieval tasks in Polish . they evaluate over 20 dense and sparse retrieval models and build sparser-dense hybrid retrievers . |
| Outcome: | The proposed framework outperforms the best available methods in 41 tasks for Polish . the proposed models outperformed the best solutions available to date . |
Towards Scalable Lightweight GUI Agents via Multi-role Orchestration (2026.findings-acl)
Copied to clipboard
Ziwei Wang, Junjie Zheng, Leyang Yang, Sheng Zhou, Xiaoxuan Tang, Fang Zhouhua, Zhiwei Liu, Dajun Chen, Yong Li, Jiajun Bu
| Challenge: | Advanced GUI agents suffer from prohibitive deployment costs on resource-constrained devices. |
| Approach: | They propose a lightweight GUI agent with GUI-specific knowledge and task scalability . LAMO-3B supports monolithic execution and MAS-style orchestration . |
| Outcome: | The proposed GUI agent LAMO-3B supports monolithic execution and MAS-style orchestration. |
Unveil: Unified Visual-Textual Integration and Distillation for Multi-modal Document Retrieval (2025.acl-long)
Copied to clipboard
| Challenge: | Document retrieval in real-world scenarios faces significant challenges due to diverse document formats and modalities. |
| Approach: | They propose a visual-textual embedding framework that integrates textual and visual features for robust document representation. |
| Outcome: | The proposed visual-textual embedding framework surpasses existing methods while preserving semantic fidelity. |
Self-Knowledge Distillation for Knowledge Graph Embedding (2024.lrec-main)
Copied to clipboard
| Challenge: | Knowledge graph embedding (KGE) is an important task for many downstream applications. |
| Approach: | They propose to use self-knowledge distillation to learn a low-dimensional model from a pre-trained high-dimensional one. |
| Outcome: | The proposed model can improve model performance while maintaining lightweight structure. |
Basic Reading Distillation (2025.acl-long)
Copied to clipboard
| Challenge: | Large language models require high computational resources which limits their deployment in real-world applications. |
| Approach: | They propose to distill large language models into smaller language models by either knowledge distillation or task distillation. |
| Outcome: | The proposed model outperforms or performs comparable to over 20x bigger LLMs on language inference benchmarks and BIG-bench tasks. |
Pru-CoT: Towards Efficient Reasoning Distillation via Pruning Chain-of-Thought (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing heuristics fail to capture global causal logic due to rigid rules and limited search spaces. |
| Approach: | They propose a framework that extracts the essential logical structure from reasoning chains. |
| Outcome: | Experiments show that Pru-CoT models generate more compact reasoning paths compared to models trained on verbose data. |
ReasMark: A Robust Watermark for Attributing LLM Reasoning Under Knowledge Distillation Attacks (2026.acl-long)
Copied to clipboard
Peizhuo Lv, Ruihua Zhou, Yunpeng Li, Ruigang Liang, Xingshuo Han, XiaoFeng Wang, Wei Dong, Yuling Liu
| Challenge: | Existing reasoning-enhanced large language models fail to provide reliable attribution of reasoning behavior once it is transferred through knowledge distillation. |
| Approach: | They propose to embed a reasoning-length gap in a model by querying a target domain and training a local student to imitate its outputs. |
| Outcome: | et al. show that ReasMark outperforms baselines while preserving task utility. |
SharVeT: Similarity-aware Parameter Sharing with Vector-based Tuning for Efficient LLM Compression (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for parameter sharing rely on naive grouping and fail to correct sharing-induced discrepancies. |
| Approach: | They propose a parameter sharing framework that performs similarity-based grouping to ensure accurate sharing and allocates parameters adaptively to preserve diversity within each group. |
| Outcome: | The proposed framework outperforms existing methods, achieving 32.1% lower perplexity and 23.3% higher few-shot reasoning accuracy. |