Mengnan Du

MemLens: Uncovering Memorization in LLMs with Activation Trajectories

AdaptiveK Sparse Autoencoders: Dynamic Sparsity Allocation for Interpretable LLM Representations

Counterfactual Visual Explanation via Causally-Guided Adversarial Steering

Explicit Column Relationship-Based Diffusion Model for High-Quality Synthetic Tabular Data Generation

Beyond Single Concept Vector: Modeling Concept Subspace in LLMs with Gaussian Distribution

Concept-Centric Token Interpretation for Vector-Quantized Generative Models

Massive Values in Self-Attention Modules are the Key to Contextual Knowledge Understanding

From Commands to Prompts: LLM-based Semantic File System for AIOS

Hallucinating LLM Could Be Creative

CORTEX: Concept-Oriented Token Explanation in Vector-Quantized Generative Model