Yichen Li

This paper investigates how Byte-Pair Encoding (BPE) tokenization causes Code LLMs to disproportionately memorize certain types of secrets, a phenomenon termed 'gibberish bias'.

The paper proposes an Augmented Model maniPulation (AugMP) strategy, utilizing graph representation learning, to effectively and stealthily manipulate federated fine-tuning of LLMs, significantly degrading global model performance while evading standard defenses.

ChainCaps introduces a novel runtime capability budgeting system that prevents 'permission laundering' in complex tool-using agents, significantly reducing attack success rates while maintaining benign functionality.

The paper introduces TimeSage-MT, a comprehensive multi-turn benchmark designed to rigorously test an LLM agent's ability to perform complex, evolving time series analysis, revealing critical gaps in current agentic reasoning.

The paper proposes DART, a test-time adaptation method that enhances zero-resource dense retrieval reranking by adaptively tuning a bilinear scoring matrix using pseudo-positive and pseudo-negative examples, achieving significant performance gains with minimal latency.