Yisen Wang

Difficult Examples Hurt Unsupervised Contrastive Learning: A Theoretical Perspective

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

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

LogiConBench: Benchmarking Logical Consistencies of LLMs

Read the Room: Video Social Reasoning with Mental-Physical Causal Chains

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

Autoregressive Models Rival Diffusion Models at ANY-ORDER Generation

MAD-Logic: Multi-Agent Debate Enhances Symbolic Translation and Reasoning

PROPGCL: Unleashing the Power of Propagation in Graph Contrastive Learning

Enhancing Logical Reasoning of Large Language Models via Phased Fine-Tuning

SelfCAD: Protecting Your Efficient Reasoning Capabilities via Self Cautious Insertion

Bootstrapping Language and Numerical Feedback for Reinforcement Learning in LLMs

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

Boosting Safety Alignment in LLMs with Response Shortcuts

Personal Tokens Matter: Towards Token-Aware Training for Personalized LLMs

Decoding Large Language Diffusion Models with Foreseeing Movement

Understanding and Leveraging Expert-level Monosemanticity in Finetuning MoE LLMs

Sensitivity as a Shield: Inducing Sensitivity to Prevent Unauthorized Model Merging

Debunking the Illusion of Backdoor Robustness in Vision Transformers

LUCID: Universal Auditing of Distilled Large Language Models

Advancing LLM Safe Alignment with Safety Representation Ranking

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

Finding and Reactivating Post-Trained LLMs' Hidden Safety Mechanisms

Leveraging Flatness to Improve Information-Theoretic Generalization Bounds for SGD

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

Projection Head is Secretly an Information Bottleneck

Identifying and Understanding Cross-Class Features in Adversarial Training

TC-MoE: Augmenting Mixture of Experts with Ternary Expert Choice

SaLoRA: Safety-Alignment Preserved Low-Rank Adaptation

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

Beyond Interpretability: The Gains of Feature Monosemanticity on Model Robustness

Output Alignment: A Top-down Approach to Length Generalization

Propagation Alone is Enough for Graph Contrastive Learning

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

Incorporating Arbitrary Matrix Group Equivariance into KANs

An Augmentation-Aware Theory for Self-Supervised Contrastive Learning

Are Synthetic Classifiers Really as Good as Real Classifiers?

EKAN: Equivariant Kolmogorov-Arnold Networks

AttnGCG: Enhancing Jailbreaking Attacks on LLMs with Attention Manipulation

Towards Reliable Backdoor Attacks on Vision Transformers