ACL 2025 Papers — Page 6
Annual Meeting of the Association for Computational Linguistics · 1699 papers
Efficient and Accurate Prompt Optimization: the Benefit of Memory in Exemplar-Guided Reflection
Cilin Yan (Beihang University), Yangyang Kang (Zhejiang University)
OptimizationComputational EfficiencyTransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: This paper proposes an Example-Guided Reflection and Memory mechanism (ERM) for efficient and precise automatic prompt optimization;
Efficient Domain Continual pretraining by Mitigating the Stability Gap
Yiduo Guo (Peking University), Dongyan Zhao (Peking University)
Domain AdaptationComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningTextBiomedical Data
🎯 What it does: During the continuous pre-training process, a 'stability gap' phenomenon was observed where model performance initially declined before recovering. Three strategies were proposed to mitigate this gap and improve domain adaptation effectiveness and computational efficiency: multi-round subset training, domain perplexity-based screening, and matching the mixing ratio of data with the original pre-training data.
Efficient Ensemble for Fine-tuning Language Models on Multiple Datasets
Dongyue Li (Northeastern University), Hongyang R. Zhang (Northeastern University)
Federated LearningComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsText
🎯 What it does: Propose an efficient low-rank adapter (LoRA/QLoRA) ensemble method for fine-tuning large language models on multiple datasets. This method uses first-order Taylor expansion to quickly estimate the fine-tuning performance of different dataset combinations, clusters tasks based on task affinity, trains a single adapter for each group, and constructs an ensemble model through gradient boosting steps and weighted averaging.
Efficient Knowledge Editing via Minimal Precomputation
Akshat Gupta, Gopala Anumanchipalli (University of Virginia)
Computational EfficiencyTransformerLarge Language ModelText
🎯 What it does: This paper proposes a method to accelerate the knowledge editing process of location-editing approaches (such as MEMIT, ROME, and EMMET) by significantly reducing the precomputation steps.
Efficient Long Context Language Model Retrieval with Compression
Minju Seo (DeepAuto), Sung Ju Hwang (DeepAuto)
RetrievalCompressionTransformerLarge Language ModelText
🎯 What it does: Propose CoLoR, a compression method for long-context language model retrieval, which trains a compression model to significantly shorten text length while maintaining retrieval accuracy.
Efficient Many-Shot In-Context Learning with Dynamic Block-Sparse Attention
Emily Xiao (Carnegie Mellon University), Amanda Bertsch (Carnegie Mellon University)
RetrievalComputational EfficiencyTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Proposes a training-free dynamic block-sparse attention framework (Dynamic Block-Sparse Attention, DBSA), which achieves efficient inference for multi-example context learning by pre-encoding the demonstration set into blocks and retrieving only the relevant block KV caches during inference.
Efficient OpAmp Adaptation for Zoom Attention to Golden Contexts
Haoyuan Wu (Chinese University of Hong Kong), Bei Yu (Chinese University of Hong Kong)
RetrievalComputational EfficiencyTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes an attention adaptation method based on the principle of operational amplifiers (OpAmp), implemented as a lightweight adapter to denoise and enhance focus on golden documents in retrieval-augmented generation (RAG) and long-text context environments for question-answering tasks.
Efficient Pretraining Data Selection for Language Models via Multi-Actor Collaboration
Tianyi Bai (Hong Kong University of Science and Technology), Conghui He (Shanghai Artificial Intelligence Laboratory)
Data-Centric LearningTransformerLarge Language ModelReinforcement LearningAgentic AIText
🎯 What it does: Propose a multi-actor collaborative data selection framework that uses various existing pre-training data screening methods (quality, domain, topic) as independent actors, dynamically integrating and updating their weights through a central console to achieve efficient dynamic screening of language model pre-training data.
Efficient Safety Alignment of Large Language Models via Preference Re-ranking and Representation-based Reward Modeling
Deng Qiyuan, Min Zhang (Harbin Institute Of Technology)
Safty and PrivacyComputational EfficiencyRepresentation LearningReinforcement Learning from Human FeedbackLarge Language ModelReinforcement LearningTextBenchmark
🎯 What it does: By assuming that target strategy sampling can be converted into preference reordering, a lightweight reward model based on internal representations is constructed, and confidence-based dynamic preference data reordering is utilized under the DPO framework to achieve safe alignment of LLMs.
Efficient Universal Goal Hijacking with Semantics-guided Prompt Organization
Yihao Huang (Nanyang Technological University), Geguang Pu (New York University)
Computational EfficiencyAdversarial AttackLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes a highly efficient and general-purpose target hijacking method called POUGH, which generates fixed suffixes using variable training prompt subsets and semantically guided prompt ordering, enabling LLMs to produce attacker-specified malicious responses when faced with arbitrary user prompts.
Efficiently Identifying Watermarked Segments in Mixed-Source Texts
Xuandong Zhao (University of California Berkeley), Lei Li (Carnegie Mellon University)
Computational EfficiencyTransformerLarge Language ModelText
🎯 What it does: This paper proposes two methods (Geometric Cover Detector GCD and Adaptive Online Locator AOL) to achieve efficient detection and localization of LLM watermark segments in long hybrid source texts.
EfficientQAT: Efficient Quantization-Aware Training for Large Language Models
Mengzhao Chen (University of Hong Kong), Ping Luo (University of Hong Kong)
Computational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Propose a new quantization-aware training framework called EfficientQAT, which can significantly reduce the memory and time required for LLM quantization training while maintaining low-bit (2/3/4-bit) accuracy.
EffiVLM-BENCH: A Comprehensive Benchmark for Evaluating Training-Free Acceleration in Large Vision-Language Models
Zekun Wang (Harbin Institute of Technology), Bing Qin (Harbin Institute of Technology)
Computational EfficiencyVision Language ModelMultimodalityBenchmark
🎯 What it does: This paper proposes EFFIVLM-BENCH, a unified evaluation framework for systematically assessing training-free acceleration methods (including token compression and parameter compression) of large vision-language models (LVLMs), covering various model architectures, tasks, and multi-dimensional metrics;
ELABORATION: A Comprehensive Benchmark on Human-LLM Competitive Programming
Xinwei Yang (Sichuan University), Wenqiang Lei (Sichuan University)
AI Code AssistantLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Proposed the ELABORATION Benchmark to evaluate the collaborative effectiveness of human-large language models (LLM) in competitive programming.
ELBA-Bench: An Efficient Learning Backdoor Attacks Benchmark for Large Language Models
Xuxu Liu (Wuhan University), Dacheng Tao (Nanyang Technological University)
Adversarial AttackTransformerSupervised Fine-TuningPrompt EngineeringTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose ELBA-Bench, a unified benchmark for backdoor attacks on large language models, covering two attack paradigms: PEFT and no Fine-tuning, and providing a standardized toolbox.
Embedding-Converter: A Unified Framework for Cross-Model Embedding Transformation
Jinsung Yoon (Google), Sercan O Arik (Google)
ClassificationRetrievalDomain AdaptationRepresentation LearningTransformerText
🎯 What it does: Proposed a unified framework called Embedding-Converter, which can efficiently convert vectors between different embedding models, avoiding the need to recompute embeddings.
Embracing Imperfection: Simulating Students with Diverse Cognitive Levels Using LLM-based Agents
Tao Wu (Zhejiang University), Fei Wu (Zhejiang University)
AI Code AssistantTransformerLarge Language ModelTextTabular
🎯 What it does: Proposes a training-free framework that constructs cognitive prototypes and uses beam search self-optimization to simulate student behaviors and solutions at different cognitive levels.
Emergent Abilities of Large Language Models under Continued Pre-training for Language Adaptation
Ahmed Elhady (University of Basque Country), Mikel Artetxe (University of Basque Country)
Domain AdaptationTransformerLarge Language ModelTextBenchmark
🎯 What it does: Investigated whether incorporating English data during language-adapted continual pre-training affects the downstream capabilities and ICL (In-Context Learning) abilities of large language models, and proposed an alternative approach without English data.
Emma-X: An Embodied Multimodal Action Model with Grounded Chain of Thought and Look-ahead Spatial Reasoning
Qi Sun (Singapore University of Technology and Design), Soujanya Poria (Nanyang Technological University)
Robotic IntelligenceLarge Language ModelVision Language ModelVision-Language-Action ModelMultimodalityChain-of-Thought
🎯 What it does: Proposed a new 7B-scale multimodal action model EMMA-X to generate low-level action sequences for robot execution;
Empathy Prediction from Diverse Perspectives
Francine Chen, Charlene C. Wu (Toyota Research Institute)
ClassificationTransformerLarge Language ModelContrastive LearningTextTabular
🎯 What it does: This paper collected the EmpathyFromPerspectives dataset and studied and implemented the PPEP model, which integrates the assessor's own perspective into empathy prediction;
Employing Discourse Coherence Enhancement to Improve Cross-Document Event and Entity Coreference Resolution
Xinyu Chen (Soochow University), Qiaoming Zhu (Soochow University)
Graph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposes the Cross-Document Discourse Coherence Enhancement (CD-DCE) task, improving cross-document event and entity coreference resolution (CDCR) performance by inserting semantically coherent text between sentences across documents.
Enabling Chatbots with Eyes and Ears: An Immersive Multimodal Conversation System for Dynamic Interactions
Jihyoung Jang (Pohang University of Science and Technology), Hyounghun Kim (Pohang University of Science and Technology)
GenerationData SynthesisRetrievalTransformerVision Language ModelImageMultimodalityRetrieval-Augmented GenerationAudio
🎯 What it does: Proposed the M3C Multimodal Multiconversation Multi-party Dialogue Dataset and trained a multimodal dialogue model that supports vision and audio.
Enabling LLM Knowledge Analysis via Extensive Materialization
Yujia Hu (ScaDS.AI & TU Dresden), Simon Razniewski (ScaDS.AI & TU Dresden)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Propose a novel method for recursive queries and result merging, constructing GPTKB—a large language model (LLM) knowledge base containing 101 million relations and nearly 3 million entities.
Energy Considerations of Large Language Model Inference and Efficiency Optimizations
Jared Fernandez (Carnegie Mellon University), Emma Strubell (Carnegie Mellon University)
Computational EfficiencyTransformerMixture of ExpertsText
🎯 What it does: This paper systematically evaluates the energy consumption during the inference process of large-scale language models and explores the impact of various software, algorithmic, and hardware optimizations on energy consumption.
English-based acoustic models perform well in the forced alignment of two English-based Pacific Creoles
Sam Passmore (Australian National University), Danielle Barth
RecognitionSupervised Fine-TuningAudio
🎯 What it does: The study compares the forced alignment performance of four acoustic models (language-specific, Pacific Creole collection, pre-trained English model, and its adapted version) in two English-based creole languages, Tok Pisin and Bislama.
Enhancing Automated Interpretability with Output-Centric Feature Descriptions
Yoav Gur-Arieh (Tel Aviv University), Mor Geva (Tel Aviv University)
Explainability and InterpretabilityTransformerLarge Language ModelAuto EncoderText
🎯 What it does: This paper proposes two output-centric feature description methods (VocabProj and TokenChange) and compares them with the traditional maximum activation sample (MaxAct) method, aiming to enhance the causal authenticity and usability of feature explanations in large language models.
Enhancing Chain-of-Thought Reasoning with Critical Representation Fine-tuning
Chenxi Huang (Zhejiang University), Jieping Ye (Alibaba Cloud Computing)
Computational EfficiencyRepresentation LearningTransformerTextBenchmarkChain-of-Thought
🎯 What it does: Propose CRFT, a parameter-efficient fine-tuning method that automatically identifies and refines key representations through information flow analysis in Chain-of-Thought reasoning to significantly improve reasoning accuracy.
Enhancing Character-Level Understanding in LLMs through Token Internal Structure Learning
Zhu Xu (Chongqing University of Posts and Telecommunications), Conglin Liu (Baidu)
RecognitionTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Propose Token Internal Position Awareness (TIPA) and Multi-Token Internal Position Awareness (MTIPA), enhancing large language models' ability to recognize character positions within subwords by training a reverse character prediction task on the subword vocabulary; applying this method to Chinese Spelling Correction (CSC) and general model fine-tuning to verify its improvements in position prediction and character-level tasks.
Enhancing Cross-Lingual Transfer through Reversible Transliteration: A Huffman-Based Approach for Low-Resource Languages
Wenhao Zhuang (Minzu University of China), Xiaobing Zhao (Minzu University of China)
CompressionDomain AdaptationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposed a reversible transcription framework based on Huffman coding, which balances compression and cross-lingual transfer;
Enhancing Event-centric News Cluster Summarization via Data Sharpening and Localization Insights
Longyin Zhang (Institute for Infocomm Research), AiTi Aw
GenerationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: This paper proposes an end-to-end event-center news clustering and summary generation framework (CLUST-MCMS), and improves the quality of multi-language, multi-domain multi-document summaries through data sharpening and localization techniques.
Enhancing Goal-oriented Proactive Dialogue Systems via Consistency Reflection and Correction
Didi Zhang (Soochow University), Qiaoming Zhu (Soochow University)
GenerationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Proposed and verified a model-agnostic two-stage Consistency Reflection and Correction (CRC) framework to enhance the consistency between generated responses and dialogue context in goal-oriented conversational systems.
Enhancing Human Evaluation in Machine Translation with Comparative Judgement
Yixiao Song (University of Massachusetts Amherst), Markus Freitag (Google)
Text
🎯 What it does: Compare traditional MQM annotations, side-by-side MQM (S×S MQM), and side-by-side relative ranking (S×S RR), investigating their impact on human evaluation consistency, error annotation consistency, system/paragraph-level ranking, and error distribution.
Enhancing Hyperbole and Metaphor Detection with Their Bidirectional Dynamic Interaction and Emotion Knowledge
Li Zheng (Wuhan University), Donghong Ji (Wuhan University)
ClassificationTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Propose an emotion-guided bidirectional dynamic interaction framework, EmoBi, which achieves joint detection of hyperbole and metaphor through emotion analysis, emotion-oriented domain mapping, and bidirectional dynamic interaction mechanisms.
Enhancing Input-Label Mapping in In-Context Learning with Contrastive Decoding
Keqin Peng, Dacheng Tao (Beihang University)
ClassificationMeta LearningTransformerContrastive LearningTextBenchmark
🎯 What it does: Propose the In-Context Contrastive Decoding (ICCD) method, which enhances the input-label mapping of LLMs in few-shot learning by contrasting positive and negative examples, thereby significantly improving multi-task NLU performance.
Enhancing Interpretable Image Classification Through LLM Agents and Conditional Concept Bottleneck Models
Yiwen Jiang (Monash University), Zongyuan Ge (Monash University)
ClassificationExplainability and InterpretabilityLarge Language ModelAgentic AIPrompt EngineeringVision Language ModelContrastive LearningImage
🎯 What it does: This paper proposes a concept agent and conditional concept bottleneck model based on LLM to achieve interpretable image classification.
Enhancing Lexicon-Based Text Embeddings with Large Language Models
Yibin Lei (University Of Amsterdam), Andrew Yates (University Of Amsterdam)
RetrievalKnowledge DistillationRepresentation LearningTransformerLarge Language ModelContrastive LearningText
🎯 What it does: Propose the LENS framework, which leverages large language models to generate dictionary-style text embeddings and enhances performance through word embedding clustering and bidirectional attention mechanisms.
Enhancing Machine Translation with Self-Supervised Preference Data
Haoxiang Sun (Shanghai Jiao Tong University), Rui Wang (Shanghai Jiao Tong University)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: Propose a self-supervised translation preference data generation framework called SSPO, which utilizes LLMs to generate erroneous annotations and corrections to create superior-inferior translation pairs, and enhances translation quality through iterative training with DPO.
Enhancing Mathematical Reasoning in LLMs by Stepwise Correction
Zhenyu Wu (Xi'an Jiaotong University), Meng Jiang (University of Notre Dame)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: Propose the STEPCO method, which gradually identifies and corrects erroneous steps in LLM-generated reasoning paths through an iterative verification-revision process, significantly improving mathematical reasoning accuracy.
Enhancing Multimodal Continual Instruction Tuning with BranchLoRA
Duzhen Zhang (Mohamed bin Zayed University of Artificial Intelligence), Jinfeng Bai (Tomorrow Advancing Life)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsVision Language ModelMultimodalityBenchmark
🎯 What it does: Propose BranchLoRA, addressing the parameter redundancy and catastrophic forgetting issues of MoELoRA in multimodal continual instruction tuning (MCIT) through asynchronous structure, flexible freezing, and task-specific routers.
Enhancing Multimodal Retrieval via Complementary Information Extraction and Alignment
Delong Zeng (Sun Yat-sen University), Ying Shen (Sun Yat-sen University)
RetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: The CIEA method is proposed for multi-modal retrieval, enhancing retrieval performance through supplementary information extraction and alignment.
Enhancing NER by Harnessing Multiple Datasets with Conditional Variational Autoencoders
Taku Oi (Toyota Technological Institute), Makoto Miwa (Toyota Technological Institute)
RecognitionAuto EncoderBiomedical Data
🎯 What it does: Fusing Conditional Variational Autoencoder (CVAE) with a span-based named entity recognition (NER) model under multiple corpora to learn shared and non-shared information between labels, thereby enhancing training effectiveness across multiple datasets.
Enhancing Neural Machine Translation Through Target Language Data: A kNN-LM Approach for Domain Adaptation
Abudurexiti Reheman (Northeastern University), JingBo Zhu (Northeastern University)
Domain AdaptationTransformerTextRetrieval-Augmented Generation
🎯 What it does: Propose the k NN-LM-NMT method, which constructs a non-parametric language model using semantically similar sentences in the target language, and combines n-gram local representations with cross-lingual retrieval similarity to enhance the domain adaptation performance of NMT.
Enhancing Open-Domain Task-Solving Capability of LLMs via Autonomous Tool Integration from GitHub
Bohan Lyu (Tsinghua University), Maosong Sun (Tsinghua University)
AI Code AssistantLarge Language ModelAgentic AITextBenchmark
🎯 What it does: Developed the OpenAct benchmark and the OpenAgent LLM agent, supporting automatic retrieval and use of tools from GitHub to address open-domain tasks.
Enhancing Retrieval Systems with Inference-Time Logical Reasoning
Felix Faltings (MIT), Yujia Bao (Accenture)
RetrievalTransformerLarge Language ModelText
🎯 What it does: Proposed a logical reasoning framework during inference, converting natural language queries into logical expressions, then vector-encoding each term and calculating cosine similarity, finally combining document scores through logical operations;
Enhancing Retrieval-Augmented Generation via Evidence Tree Search
Hao Sun (Peking University), Dawei Yin (Baidu Inc)
GenerationRetrievalComputational EfficiencyTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: Proposes the ETS framework, modeling evidence retrieval as a tree expansion process, and efficiently mining multi-sentence evidence using MCTS combined with early-stopped Beam Search.
Enhancing Safe and Controllable Protein Generation via Knowledge Preference Optimization
Yuhao Wang (Zhejiang University), Huajun Chen (Zhejiang University)
Drug DiscoveryGraph Neural NetworkTransformerLarge Language ModelReinforcement LearningGraphBiomedical Data
🎯 What it does: Propose the KPO framework, combining a protein safety knowledge graph with reinforcement learning to achieve safe alignment for protein language models.
Enhancing Spoken Discourse Modeling in Language Models Using Gestural Cues
Varsha Suresh (Saarland University), Vera Demberg (Saarland University)
Representation LearningTransformerLarge Language ModelVision-Language-Action ModelAuto EncoderTextMultimodalityPoint Cloud
🎯 What it does: This paper investigates integrating gestures (non-verbal cues) with language models to enhance discourse modeling capabilities in spoken language.
Enhancing Text Editing for Grammatical Error Correction: Arabic as a Case Study
Bashar Alhafni (Mohamed bin Zayed University of Artificial Intelligence), Nizar Habash (New York University Abu Dhabi)
RestorationData-Centric LearningTransformerLarge Language ModelText
🎯 What it does: This paper proposes a data-driven text editing framework for grammar error correction in Arabic (including Modern Standard Arabic and dialects), eliminating the need for language-specific edit tags;
Enhancing Transformers for Generalizable First-Order Logical Entailment
Tianshi Zheng (HKUST), Jianxin Li (Beihang University)
Representation LearningTransformerGraph
🎯 What it does: Study the generalization ability of Transformer under parameterized knowledge for first-order logic reasoning, and propose a logic-aware Transformer architecture (TEGA) to significantly enhance reasoning performance.
Enhancing Unsupervised Sentence Embeddings via Knowledge-Driven Data Augmentation and Gaussian-Decayed Contrastive Learning
Peichao Lai (Peking University), Bin Cui (Peking University)
Representation LearningData-Centric LearningTransformerLarge Language ModelContrastive LearningText
🎯 What it does: Propose a pipeline-based data augmentation method combining knowledge graphs and LLMs, and introduce the Gaussian-decayed Gradient-assisted Contrastive Sentence Embedding (GCSE) model to improve the quality of unsupervised sentence embeddings.
Enough Coin Flips Can Make LLMs Act Bayesian
Ritwik Gupta (University of California Berkeley), David M. Chan (University of California Berkeley)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper investigates whether large language models (LLMs) perform Bayesian inference in in-context learning (ICL) by introducing contextual examples in a biased coin flip experiment, and analyzes the model's prior, update process, attention effects, and scaling effects.
Ensemble Watermarks for Large Language Models
Georg Niess (Graz University of Technology), Roman Kern (Know Center Research GmbH)
Safty and PrivacyData-Centric LearningTransformerLarge Language ModelText
🎯 What it does: This study proposes a multi-feature ensemble watermarking scheme for large language models (LLMs), integrating acrostic, sensorimotor norms, and red-green lists. It achieves embedding and detection through dynamic weighting of model logits.
Entailed Between the Lines: Incorporating Implication into NLI
Shreya Havaldar (University of Pennsylvania), Alex Fabrikant (Google Deepmind)
ClassificationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Proposed and constructed the Implied NLI (INLI) dataset, extending traditional NLI tasks to distinguish between explicit and implied entailment.
Entailment-Preserving First-order Logic Representations in Natural Language Entailment
Jinu Lee (University of Illinois Urbana-Champaign), Julia Hockenmaier (University of Illinois Urbana-Champaign)
Representation LearningTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: Proposed and implemented the Entailment-Preserving First-Order Logic (EPF) task, constructed a reference-free evaluation metric (EPR series), and developed a training method based on iterative learning-to-rank, using natural language reasoning labels as verifiable rewards to enhance the reasoning fidelity of NL→FOL translators.
EpMAN: Episodic Memory AttentioN for Generalizing to Longer Contexts
Subhajit Chaudhury (IBM), Matthew Riemer (IBM)
RetrievalTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: Propose EpMAN (an attention mechanism based on episodic memory), helping large language models better retrieve and utilize important information in long contexts
EPO: Explicit Policy Optimization for Strategic Reasoning in LLMs via Reinforcement Learning
Xiaoqian Liu (University of Chinese Academy of Sciences), Junge Zhang (Chinese Academy of Sciences)
OptimizationTransformerLarge Language ModelReinforcement LearningAgentic AIPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Propose the EPO framework, which trains a specialized strategy generation LLM and uses it as an external policy guide to collaborate with any LLM agent, enhancing the ability to achieve long-term goals in dynamic environments.
Error Comparison Optimization for Large Language Models on Aspect-Based Sentiment Analysis
Qianlong Wang (Harbin Institute of Technology), Ruifeng Xu (Harbin Institute of Technology)
ClassificationTransformerSupervised Fine-TuningText
🎯 What it does: By constructing error comparison pairs on large language models and introducing a calibration loss, the Aspect-Based Sentiment Analysis task is refined through fine-tuning, enabling the model to perceive and reduce errors of varying degrees.
Error-driven Data-efficient Large Multimodal Model Tuning
Barry Menglong Yao (UC Davis), Lifu Huang (UC Davis)
Knowledge DistillationData-Centric LearningSupervised Fine-TuningPrompt EngineeringTextMultimodalityRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposed an error-driven, data-efficient tuning framework that uses a student-teacher model to identify defects in multi-modal models and retrieves targeted training samples from task-agnostic datasets to quickly adapt to new tasks;
ERU-KG: Efficient Reference-aligned Unsupervised Keyphrase Generation
Lam Thanh Do (University of Illinois Urbana-Champaign), Kevin Chen-Chuan Chang (University of Illinois Urbana-Champaign)
GenerationRetrievalComputational EfficiencyRepresentation LearningData-Centric LearningTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose an unsupervised key phrase generation model, ERU-KG, which can simultaneously predict both existing and missing key phrases and switch to extraction mode via hyperparameters.
EscapeBench: Towards Advancing Creative Intelligence of Language Model Agents
Cheng Qian (University of Illinois Urbana-Champaign), Heng Ji (University of Illinois Urbana-Champaign)
Large Language ModelAgentic AITextBenchmarkChain-of-Thought
🎯 What it does: Propose EscapeBench to evaluate the creative reasoning of LM agents, and build the EscapeAgent model to enhance tool creative usage and implicit goal recognition.
Establishing Trustworthy LLM Evaluation via Shortcut Neuron Analysis
Kejian Zhu (University of Chinese Academy of Sciences), Jun Zhao (University of Chinese Academy of Sciences)
Explainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelTextBenchmark
🎯 What it does: Proposes a neuron-level shortcut analysis and repair method to eliminate overfitting of large models on contaminated data, thereby enhancing assessment credibility.
Estimating Privacy Leakage of Augmented Contextual Knowledge in Language Models
James Flemings (University of Southern California), Murali Annavaram (University of Southern California)
Safty and PrivacyTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper investigates the privacy leakage issue in language models when using enhanced context, and proposes a metric called 'context influence' to quantify the privacy leakage of contextual knowledge.
ETF: An Entity Tracing Framework for Hallucination Detection in Code Summaries
Kishan Maharaj (University of Texas at Austin), Pushpak Bhattacharyya (Indian Institute of Technology Bombay)
ClassificationAnomaly DetectionAI Code AssistantTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper proposes an Entity Tracking Framework (ETF) to detect hallucinations in code summaries and creates a specialized dataset called CodeSumEval.
Evaluating Design Decisions for Dual Encoder-based Entity Disambiguation
Susanna Rücker (Humboldt-Universität zu Berlin), Alan Akbik (Humboldt-Universität zu Berlin)
RetrievalRepresentation LearningTransformerSupervised Fine-TuningContrastive LearningText
🎯 What it does: This paper systematically evaluates critical design decisions of dual encoders in entity disambiguation, and proposes the VERBALIZED system based on these insights, further exploring iterative prediction strategies.
Evaluating Language Models as Synthetic Data Generators
Seungone Kim (Carnegie Mellon University), Graham Neubig (Carnegie Mellon University)
Data SynthesisTransformerLarge Language ModelTextBenchmark
🎯 What it does: Constructed the AGORABENCH benchmark to systematically evaluate the ability of different language models in generating synthetic data.
Evaluating Lexical Proficiency in Neural Language Models
Cristiano Ciaccio (Istituto di Linguistica Computazionale 'Antonio Zampolli' (CNR-ILC)), Felice Dell’Orletta (Istituto di Linguistica Computazionale 'Antonio Zampolli' (CNR-ILC))
GenerationTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Proposes a unified evaluation framework to assess the lexical capabilities of Transformer-based language models in terms of vocabulary generation, definition, and contextual usage, with experiments conducted on Italian.
Evaluating LLMs for Portuguese Sentence Simplification with Linguistic Insights
Arthur Mariano Rocha De Azevedo Scalercio (Universidade Federal Fluminense), Aline Paes (Universidade de São Paulo)
GenerationTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Evaluated the one-shot performance of 26 large language models on Portuguese sentence simplification tasks and released a new GovLang-BR simplification corpus.
Evaluating Multimodal Language Models as Visual Assistants for Visually Impaired Users
Antonia Karamolegkou (University of Copenhagen), Anders Søgaard (University of Copenhagen)
RecognitionObject DetectionTransformerLarge Language ModelVision Language ModelImageVideoTextMultimodalityBenchmark
🎯 What it does: This paper systematically evaluates the effectiveness of multimodal large language models (MLLM) in visually assisted blind individuals, proposes user surveys and a five-task framework, and constructs a related evaluation dataset;
Evaluating Multimodal Large Language Models on Video Captioning via Monte Carlo Tree Search
Linhao Yu (Tianjin University), Deyi Xiong (Tianjin University)
GenerationLarge Language ModelSupervised Fine-TuningVision Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: Use Monte Carlo Tree Search (MCTS) to automatically generate key points for video captions in multimodal large language models, constructing a benchmark to evaluate their video understanding capabilities.
Evaluating Personalized Tool-Augmented LLMs from the Perspectives of Personalization and Proactivity
Yupu Hao (Chinese Academy of Sciences), Jun Zhao (University of Chinese Academy of Sciences)
TransformerLarge Language ModelSupervised Fine-TuningAgentic AITextBenchmark
🎯 What it does: Proposed the ETAPP benchmark for evaluating large language models (LLMs) in personalized tool calling, and evaluated it through a sandbox environment with 800 multi-user scenario test cases;
Evaluating Sequence Labeling on the basis of Information Theory
Enrique Amigó (UNED), Jorge Carrillo-de-Albornoz (UNED)
ClassificationText
🎯 What it does: A set of formal attributes for evaluating sequence labeling tasks is proposed, along with a novel information-theoretic metric called SL-ICM that simultaneously satisfies all attributes;
Evaluating the Evaluation of Diversity in Commonsense Generation
Tianhui Zhang (University of Liverpool), Danushka Bollegala (University of Liverpool)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper systematically conducts a meta-evaluation of various metrics used to measure diversity in the commonsense generation (GCR) task, constructs a diversity-annotated dataset based on large language model (LLM) annotations, and evaluates the reliability of the metrics on three public datasets.
Evaluating Theory of (an uncertain) Mind: Predicting the Uncertain Beliefs of Others from Conversational Cues
Anthony Sicilia (Northeastern University), Malihe Alikhani (Northeastern University)
TransformerLarge Language ModelSupervised Fine-TuningTextChain-of-Thought
🎯 What it does: This study proposes and implements a theory of mind (ToM) task based on dialogue prediction, aiming to enable language models to predict others' subjective uncertainty about a belief through linguistic cues in dialogue (i.e., 'false uncertainty').
Evaluating Visual and Cultural Interpretation: The K-Viscuit Benchmark with Human-VLM Collaboration
ChaeHun Park (KAIST AI), Jaegul Choo (KAIST AI)
Explainability and InterpretabilityPrompt EngineeringVision Language ModelImageTextMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Proposed a semi-automated multi-modal cultural evaluation framework that leverages collaboration between humans and large vision-language models (VLMs) to generate multiple-choice questions and answers, with validation by native speakers; under this framework, constructed the K-Viscuit benchmark dataset focusing on Korean culture;
Evaluation Agent: Efficient and Promptable Evaluation Framework for Visual Generative Models
Fan Zhang (Shanghai Artificial Intelligence Laboratory), Ziwei Liu (Nanyang Technological University)
GenerationExplainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelAgentic AIPrompt EngineeringVision Language ModelImageVideoBenchmark
🎯 What it does: Proposed a framework called Evaluation Agent, which utilizes LLM agents to dynamically plan, generate prompts, and evaluate visual generation models through multi-round interactions, achieving an efficient, customizable, and interpretable evaluation process.
Evaluation of LLM Vulnerabilities to Being Misused for Personalized Disinformation Generation
Aneta Zugecova (Kempelen Institute of Intelligent Technologies), Matúš Mesarčík (Kempelen Institute of Intelligent Technologies)
GenerationSafty and PrivacyTransformerLarge Language ModelText
🎯 What it does: This paper constructs the PerDisNews dataset to systematically evaluate the vulnerability of large language models (LLMs) when generating personalized fake news, the quality of personalization, and the behavior of safety filters, and verifies that multi-model meta-assessment methods can be used to assess the degree of personalization.
EventRAG: Enhancing LLM Generation with Event Knowledge Graphs
Zairun Yang (Zhejiang University), Qiang Zhang (Zhejiang University)
GenerationRetrievalGraph Neural NetworkLarge Language ModelAgentic AITextGraphRetrieval-Augmented Generation
🎯 What it does: Propose an event knowledge graph (Event Knowledge Graph, EKG)-based retrieval-augmented generation framework called EventRAG, which first extracts and fuses events and entities from multiple documents to construct a structured event graph; subsequently, it utilizes agent-based multi-step retrieval and reasoning to fully leverage the temporal and logical relationships of events during generation.
EvolveBench: A Comprehensive Benchmark for Assessing Temporal Awareness in LLMs on Evolving Knowledge
Zhiyuan Zhu (Shanghai Jiao Tong University), Yu Wang (Shanghai Jiao Tong University)
Large Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Proposed the EvolveBench benchmark to systematically evaluate the capabilities of large language models (LLMs) in temporal evolution knowledge, covering five dimensions: cognition, temporal disorientation awareness, credibility, temporal understanding, and reasoning.
EvoWiki: Evaluating LLMs on Evolving Knowledge
Wei Tang, Yong Liao
TransformerLarge Language ModelSupervised Fine-TuningTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: This study proposes EvoWiki, a dynamic and automatically updatable benchmark for evaluating large language models (LLM) in their ability to utilize knowledge that evolves over time.
Exclusion of Thought: Mitigating Cognitive Load in Large Language Models for Enhanced Reasoning in Multiple-Choice Tasks
Qihang Fu (Guizhou University), Lintao Long (Guizhou University)
Explainability and InterpretabilityTransformerPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Propose a prompt framework called EoT based on the 'gradual elimination' thinking, helping large language models progressively eliminate incorrect answers in multiple-choice questions, reducing cognitive load and improving reasoning accuracy.
ExpeTrans: LLMs Are Experiential Transfer Learners
Jinglong Gao (Harbin Institute of Technology), Ting Liu (Harbin Institute of Technology)
ClassificationMeta LearningTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes ExpeTrans—a self-autonomous experience transfer framework based on LLM, capable of automatically extracting task experience from existing labeled data and transferring these experiences to new tasks during inference to enhance performance.
Explaining Matters: Leveraging Definitions and Semantic Expansion for Sexism Detection
Sahrish Khan (University of Warwick), Gabriele Pergola (University of Warwick)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringMixture of ExpertsTextChain-of-Thought
🎯 What it does: Data augmentation for online sexist text using category definitions and semantic expansion, achieving fine-grained detection through multi-model voting with Mistral-7B as a fallback model.
Explicit and Implicit Data Augmentation for Social Event Detection
Congbo Ma (Macquarie University), Preslav Nakov (Macquarie University)
ClassificationData SynthesisGraph Neural NetworkLarge Language ModelTextGraph
🎯 What it does: Propose a dual data augmentation framework SED-Aug, combining LLM-based text augmentation with structural fusion embedding perturbation in the feature space to enhance social event detection performance;
Exploiting Contextual Knowledge in LLMs through \mathcal{V}-usable Information based Layer Enhancement
Xiaowei Yuan (Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences), Kang Liu (Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences)
Representation LearningTransformerLarge Language ModelText
🎯 What it does: Propose the Context-aware Layer Enhancement (CaLE) method, which improves the contextual authenticity of LLMs by amplifying or using residual connections for contextual information within internal layers.
Exploiting the Shadows: Unveiling Privacy Leaks through Lower-Ranked Tokens in Large Language Models
Yuan Zhou (Purdue University), Xiangyu Zhang (Purdue University)
Safty and PrivacyLarge Language ModelText
🎯 What it does: This paper proposes a novel attack method based on the leakage of low-rank output tokens from LLMs, exploiting the model's covert leakage of private documents or PII information when generating low-ranked tokens.
ExploraCoder: Advancing Code Generation for Multiple Unseen APIs via Planning and Chained Exploration
Yunkun Wang (Zhejiang University), Shuiguang Deng (Zhejiang University)
AI Code AssistantTransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose a training-agnostic framework called ExploraCoder, enabling LLMs to invoke multiple unseen APIs in code;
Exploring Compositional Generalization of Multimodal LLMs for Medical Imaging
Zhenyang Cai (Chinese University of Hong Kong), Benyou Wang (Chinese University of Hong Kong)
TransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodalityBiomedical DataBenchmark
🎯 What it does: Construct Med-MAT, a medical image VQA dataset based on MAT-Triplet (Modality-Anatomy-Task), and explore its combination reasoning (CG) and its impact on multi-task generalization using multimodal large language models (MLLM).
Exploring Explanations Improves the Robustness of In-Context Learning
Ukyo Honda (CyberAgent), Tatsushi Oka (Keio University)
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes an improved context learning method called X2-ICL, which enhances the robustness to out-of-distribution data by generating and comparing explanations across all possible labels using the model.
Exploring Forgetting in Large Language Model Pre-Training
Chonghua Liao (Tsinghua University), Zhanhui Kang (Tsinghua University)
OptimizationTransformerLarge Language ModelText
🎯 What it does: Systematically investigate the catastrophic forgetting phenomenon during the pre-training phase of LLMs and propose entity-based evaluation metrics
Exploring How Generative MLLMs Perceive More Than CLIP with the Same Vision Encoder
Siting Li (University of Washington), Simon Shaolei Du (University of Washington)
RecognitionTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelContrastive LearningImageMultimodalityBenchmark
🎯 What it does: Compare the visual reasoning capabilities of CLIP and generative multimodal large language models (e.g., LLaVA, Phi-3V, LLaMA-3-V) under the same visual encoder, and analyze the sources of differences through a series of controlled experiments.
Exploring LLMs’ Ability to Spontaneously and Conditionally Modify Moral Expressions through Text Manipulation
Candida Maria Greco (University of Calabria), Andrea Tagarelli (University of Calabria)
GenerationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Investigate the spontaneous and conditional modification capabilities of LLMs in text editing tasks (e.g., rewriting, continuation), quantitatively evaluating changes in moral dimensions of texts before and after modification using Moral Foundations Theory.
Exploring Multimodal Relation Extraction of Hierarchical Tabular Data with Multi-task Learning
Xinyu Zhang (Southeast University), Xiaolin Fang (Southeast University)
TransformerLarge Language ModelMultimodalityTabularChain-of-Thought
🎯 What it does: The study addresses multi-relational extraction for hierarchical tables in a multi-modal environment, proposing a multi-task learning framework that computes relations and integrates semantic relations.
Exploring Persona Sentiment Sensitivity in Personalized Dialogue Generation
Yonghyun Jun (Chung-Ang University), Hwanhee Lee (Chung-Ang University)
GenerationTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Investigated the impact of users' personality emotional polarity on personalized dialogue generation, and proposed a dialogue generation framework based on emotional polarity adjustment.
Exploring the Impact of Instruction-Tuning on LLM’s Susceptibility to Misinformation
Kyubeen Han (Konkuk University), Harksoo Kim (ETRI)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Investigated the susceptibility of large language models to misinformation provided by users when instruction-tuning is applied, and systematically evaluated the impact of different prompt structures and misinformation lengths on model behavior.
Exploring the Potential of LLMs as Personalized Assistants: Dataset, Evaluation, and Analysis
Jisoo Mok (Seoul National University), Sungroh Yoon (Seoul National University)
Data SynthesisTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose the HiCUPID dataset and evaluation framework for training and assessing the capability of LLMs as personalized assistants, and provide the Llama-3.2 agent evaluator.
Exposing the Achilles’ Heel: Evaluating LLMs Ability to Handle Mistakes in Mathematical Reasoning
Joykirat Singh (Microsoft Research), Vibhav Vineet (Microsoft Research)
Explainability and InterpretabilityTransformerLarge Language ModelTextBenchmark
🎯 What it does: Constructed a mathematical word problem (MWP) dataset containing erroneous reasoning steps, and assessed the performance of large language models (LLMs) in detecting and correcting reasoning errors.
Extending Complex Logical Queries on Uncertain Knowledge Graphs
Weizhi Fei (Princeton University), Yangqiu Song (Hong Kong University of Science and Technology)
Graph Neural NetworkGraph
🎯 What it does: The study addresses answering soft logic queries on uncertain knowledge graphs, proposing a neuro-symbolic method named SRC that combines forward reasoning with backward calibration.
Extending LLM Context Window with Adaptive Grouped Positional Encoding: A Training-Free Method
Xinhao Xu (Tsinghua University), Guiguang Ding (Tsinghua University)
RetrievalRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Propose a training-free, plug-and-play adaptive grouped positional encoding method called AdaGroPE for expanding the context window of large language models.
F5-TTS: A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matching
Yushen Chen (Shanghai Jiao Tong University), Xie Chen (Geely Automobile Research Institute Company Ltd.)
GenerationConvolutional Neural NetworkTransformerDiffusion modelFlow-based ModelTextOrdinary Differential EquationAudio
🎯 What it does: Proposed F5-TTS, a fully non-autoregressive text-to-speech system based on flow matching and Diffusion Transformer; fills character sequence alignment length to avoid complex duration models, phoneme alignment, text encoders, etc.
FACT-AUDIT: An Adaptive Multi-Agent Framework for Dynamic Fact-Checking Evaluation of Large Language Models
Hongzhan Lin (Hong Kong Baptist University), Tat-Seng Chua (National University of Singapore)
Explainability and InterpretabilityData-Centric LearningLarge Language ModelAgentic AITextBenchmark
🎯 What it does: Proposed FACT-AUDIT, a multi-agent adaptive evaluation framework for dynamically detecting the fact-checking capabilities of large language models (LLMs), while simultaneously assessing the evidence provided by the models.