Jingyi Cui

On the Limits of Sparse Autoencoders: A Theoretical Framework and Reweighted Remedy

Difficult Examples Hurt Unsupervised Contrastive Learning: A Theoretical Perspective

Target-Aware Normalized Distillation: A Principled Framework for Robust Knowledge Transfer

LUCID: Universal Auditing of Distilled Large Language Models

Beyond Interpretability: The Gains of Feature Monosemanticity on Model Robustness

A Unified Theoretical Framework for Understanding Difficult-to-learn Examples in Contrastive Learning

An Augmentation-Aware Theory for Self-Supervised Contrastive Learning