research
our papers organized by topic, preprints first, then published work. The same papers listed by year are on the publications page.
Click a topic to expand its papers.
Large Language Models
Data Selection, Curriculum & Synthetic DataPost-training & ReasoningHuman–AI Alignment & Reward HackingGeneralization, Memorization & ArchitecturePrompting & Test-time OptimizationIn-context LearningEfficient DecodingRetrieval & EmbeddingsSafety & JailbreakingBenchmarks & Evaluation
Data Selection, Curriculum & Synthetic Data
-
- Can LLMs Speak For Diverse People? Tuning LLMs via Debate to Generate Controllable Controversial StatementsIn Annual Meeting of the Association for Computational Linguistics (ACL) Findings, 2024
- Selective Reflection-Tuning: Student-Selected Data Recycling for LLM Instruction-TuningIn Annual Meeting of the Association for Computational Linguistics (ACL) Findings, 2024
- Alpagasus: Training a Better Alpaca Model with Fewer DataIn International Conference on Learning Representations (ICLR), 2024
Post-training & Reasoning
- Wait, We Don’t Need to" Wait"! Removing Thinking Tokens Improves Reasoning EfficiencyIn The 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP) Findings, 2025
-
Human–AI Alignment & Reward Hacking
-
-
- From Lists to Emojis: How format bias affects model alignmentIn The 63rd Annual Meeting of the Association for Computational Linguistics (ACL), 2025
Generalization, Memorization & Architecture
- PreprintDoing More with Less: Monitoring Generalization During Pretraining via Virtual MoEUnder review, 2026
Prompting & Test-time Optimization
- When Do You Need Chain-of-Thought Prompting for ChatGPT?arXiv preprint arXiv:2304.03262, 2023
In-context Learning
-
- How Many Demonstrations Do You Need for In-context Learning?In The 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP) Findings, 2023
Efficient Decoding
Retrieval & Embeddings
- Large Language Models are Strong Zero-Shot RetrieverIn Annual Meeting of the Association for Computational Linguistics (ACL) Findings, 2024
Safety & Jailbreaking
- DrAttack: Prompt Decomposition and Reconstruction Makes Powerful LLM JailbreakersIn The 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP) Findings, 2024
Benchmarks & Evaluation
Multimodal Models
Architectures & Unified ModelsPost-training & AlignmentBenchmarks, Hallucination & EvaluationTest-time OptimizationDiffusion Models
Architectures & Unified Models
Post-training & Alignment
- Wait, We Don’t Need to" Wait"! Removing Thinking Tokens Improves Reasoning EfficiencyIn The 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP) Findings, 2025
- Enhancing Visual-Language Modality Alignment in Large Vision Language Models via Self-ImprovementIn Annual Conference of the Nations of the Americas Chapter of the ACL (NAACL) Findings, 2025
Benchmarks, Hallucination & Evaluation
- AUTOHALLUSION: Automatic Generation of Hallucination Benchmarks for Vision-Language ModelsIn The 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP) Findings, 2024
Test-time Optimization
Diffusion Models
- When and How do negative prompts take effect?In European Conference on Computer Vision (ECCV), 2024
Agents & World Models
Agents, World Models & Embodied AI
- Preprint
- CAIS
-
- Click, Type, Repeat: A Comprehensive Survey on GUI AgentsIn The 63rd Annual Meeting of the Association for Computational Linguistics (ACL) Findings, 2025
-
Reinforcement Learning
- TPAMIFalse Correlation Reduction for Offline Reinforcement LearningIEEE Transactions on Pattern Analysis and Machine Intelligence, 2023
Learning Paradigms
Curriculum Learning & Data SelectionData Augmentation & SynthesisContinual LearningTransfer, Multi-task & Meta-learningFederated & Collaborative Learning
Curriculum Learning & Data Selection
- Robust Curriculum Learning: from clean label detection to noisy label self-correctionIn International Conference on Learning Representations (ICLR), 2021
- Minimax Curriculum Learning: Machine Teaching with Desirable Difficulties and Scheduled DiversityIn Sixth International Conference on Learning Representations (ICLR), Vancouver, BC, Canada, 2018
- TPAMITime-Consistency Curriculum for Learning from Instance-Dependent Noisy LabelsIEEE Transactions on Pattern Analysis and Machine Intelligence, 2024
- TNNLSData Subset Selection with Imperfect Multiple LabelsIEEE Transactions on Neural Networks and Learning Systems, 2018
Data Augmentation & Synthesis
- AUTOHALLUSION: Automatic Generation of Hallucination Benchmarks for Vision-Language ModelsIn The 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP) Findings, 2024
- Subclass-balancing Contrastive Learning for Long-tailed RecognitionIn International Conference on Computer Vision (ICCV), 2023
Continual Learning
Transfer, Multi-task & Meta-learning
Federated & Collaborative Learning
- GPFedRec: Graph-Guided Personalization for Federated RecommendationIn 30th SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), 2024
- Dual Personalization on Federated RecommendationIn International Joint Conference on Artificial Intelligence (IJCAI), 2023
- Federated Learning from Pre-Trained Models: A Contrastive Learning ApproachIn Advances in Neural Information Processing Systems 36 (NeurIPS), 2022
- Personalized Federated Learning With Structural InformationOralIn International Joint Conference on Artificial Intelligence (IJCAI), 2022
- Learning to Collaborate in Decentralized Learning of Personalized ModelsIn IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022
- WWWJ
Foundations & Earlier Work
Attention & NLP (pre-LLM)Matrix Factorization & DisentanglementCompressed Sensing & Sparse LearningOptimization
Attention & NLP (pre-LLM)
- Merging Experts into One: Improving Computational Efficiency of Mixture of ExpertsIn The 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP), 2023
-
- Token Dropping for Efficient BERT PretrainingIn Annual Meeting of the Association for Computational Linguistics (ACL), 2022
-
- TIP
Matrix Factorization & Disentanglement
Compressed Sensing & Sparse Learning
- TNNLSMinimizing Nearest Neighbor Classification Error for Nonparametric Dimension ReductionIEEE Transactions on Neural Networks and Learning Systems, 2014