💼 Experience

Interpretability of Large Language Models

Research Assistant — Yale University · Jul 2025 – Sep 2026

Developed an end-to-end LLM interpretability pipeline integrating cross-layer transcoder features, attribution analysis, and circuit tracing. Designed sparse feature representations achieving 40× higher sparsity while preserving model performance, and conducted controlled intervention experiments to analyze model behavior.

Blog: What Makes a Language Model Keep Reasoning · [source]

Image Restoration and Inverse Problems

Researcher — Nanjing University · Oct 2024 – Aug 2025

Developed the official open-source implementation of Reverse Convolution, which has received over 100 GitHub stars. Implemented a reusable closed-form reverse-convolution operator and a modular evaluation pipeline for image restoration and deblurring.

Publication: Reverse Convolution and Its Applications to Image Restoration — ICCV 2025, Co-first Author.

Computer Vision and Facial Privacy Protection

Undergraduate Researcher — Nanjing University · Aug 2024 – Apr 2025

Developed a hierarchical adversarial optimization pipeline to generate protective perturbations against customized text-to-image synthesis, including reward-guided optimization and evaluation workflows.

Preprint: Anti-Aesthetics: Protecting Facial Privacy against Customized Text-to-Image Synthesis — arXiv, 2025 (withdrawn).