Wei Lin

The paper introduces CI-Work, a benchmark demonstrating that current enterprise LLM agents frequently leak sensitive information while performing tasks, suggesting that privacy protection requires architectural changes rather than just larger models.

The paper proposes MAS-SZZ, a multi-agentic algorithm that significantly improves the identification of the earliest commit introducing a software vulnerability by combining root cause analysis with structured backward tracing.

The paper proposes CFE-PPAR, the first compression-friendly encryption method for privacy-preserving action recognition, allowing video transformers to recognize actions directly from compressed, encrypted videos.

ZipRL introduces an adaptive context compression framework that significantly improves the performance and efficiency of LLMs in complex, multi-turn agent tasks by combining multi-granularity compression with Hindsight Response Replay.

This paper proposes two horizon-control strategies, Progressive OPD (POPD) and Truncated OPD (TOPD), demonstrating that full rollouts are often unnecessary for On-Policy Distillation, leading to significant improvements in training efficiency.