Yifei Wang

When More is Less: Understanding Chain-of-Thought Length in LLMs

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

CSRv2: Unlocking Ultra-Sparse Embeddings

SSL4RL: Revisiting Self-supervised Learning as Intrinsic Reward for Visual-Language Reasoning

The Chosen Few: Sparse Adaptation for Large Models

Autoregressive Models Rival Diffusion Models at ANY-ORDER Generation

Scaling Attention via Feature Sparsity

PROPGCL: Unleashing the Power of Propagation in Graph Contrastive Learning

SAE as a Crystal Ball: Interpretable Features Predict Cross-domain Transferability of LLMs without Training

Route Experts by Sequence, Not by Token

Bootstrapping Language and Numerical Feedback for Reinforcement Learning in LLMs

Debunking the Illusion of Backdoor Robustness in Vision Transformers

MULTI-FACETED MULTIMODAL MONOSEMANTICITY

Beyond Matryoshka: Revisiting Sparse Coding for Adaptive Representation

$G1$: Teaching LLMs to Reason on Graphs with Reinforcement Learning

A Signed Graph Approach to Understanding and Mitigating Oversmoothing

What is Wrong with Perplexity for Long-context Language Modeling?

Projection Head is Secretly an Information Bottleneck

On the Emergence of Position Bias in Transformers

Next Semantic Scale Prediction via Hierarchical Diffusion Language Models

Language Ranker: A Lightweight Ranking framework for LLM Decoding

Can In-context Learning Really Generalize to Out-of-distribution Tasks?

Long-Short Alignment for Effective Long-Context Modeling in LLMs

Rethinking Invariance in In-context Learning

Tool Decoding: A Plug-and-Play Approach to Enhancing Language Models for Tool Usage

Beyond Interpretability: The Gains of Feature Monosemanticity on Model Robustness

Output Alignment: A Top-down Approach to Length Generalization

Oversmoothing as Loss of Sign: Towards Structural Balance in Graph Neural Networks

Propagation Alone is Enough for Graph Contrastive Learning

Are Synthetic Classifiers Really as Good as Real Classifiers?

Towards Reliable Backdoor Attacks on Vision Transformers