趙 羽風

Mechanism of Task-oriented Information Removal in In-context Learning 

The Fourteenth International Conference on Learning Representations (ICLR) 2026年5月  査読有り

Localizing Task Recognition and Task Learning in In-Context Learning via Attention Head Analysis 

The Fourteenth International Conference on Learning Representations (ICLR) 2026年5月  査読有り

Task Vectors, Learned Not Extracted: Performance Gains and Mechanistic Insight 

The Fourteenth International Conference on Learning Representations (ICLR) 2026年5月  査読有り

Unifying Attention Heads and Task Vectors via Hidden State Geometry in In-Context Learning 

Neural Information Processing Systems (NeurIPS) 2025年9月  査読有り

Mechanistic Fine-tuning for In-context Learning 

BlackboxNLP Workshop @ EMNLP 2025 2025年5月  査読有り

Revisiting In-context Learning Inference Circuit in Large Language Models 

The Thirteenth International Conference on Learning Representations (ICLR) 2025年4月  査読有り

Token-based Decision Criteria Are Suboptimal in In-context Learning 

In Proceedings of the 2025 Annual Conference of the Nations of the Americas Chapter of the ACL (NAACL.main) 2025年4月  査読有り

Understanding Token Probability Encoding in Output Embeddings 

In Proceedings of the 2025 Conference on Computational Linguistics (COLING) 2025年1月  査読有り