Ruoxi Jia

Can Small Training Runs Reliably Guide Data Curation? Rethinking Proxy-Model Practice

Quagmires in SFT-RL Post-Training: When High SFT Scores Mislead and What to Use Instead

Adversarial Déjà Vu: Jailbreak Dictionary Learning for Stronger Generalization to Unseen Attacks

Safety at One Shot: Patching Fine-Tuned LLMs with A Single Instance

CASPO: Confidence-aware Step-wise Preference Optimization for Reliable Reasoning in Large Language Models

Inference-Time Personalized Safety Control via Paired Difference-in-Means Intervention

Data Valuation and Selection in a Federated Model Marketplace

ARTS: Alleviating Hallucinations in Large Vision–Language Models via Redundancy-Aware Token Selection

AdaDeDup: Adaptive Hybrid Data Pruning for Efficient Object Detection Training

Distilling Reasoning into Student LLMs: Local Naturalness for Selecting Teacher Data

Capturing the Temporal Dependence of Training Data Influence

Data Shapley in One Training Run

AIR-BENCH 2024: A Safety Benchmark based on Regulation and Policies Specified Risk Categories

LLM Can be a Dangerous Persuader: Empirical Study of Persuasion Safety in Large Language Models

SORRY-Bench: Systematically Evaluating Large Language Model Safety Refusal

Data-Centric Human Preference with Rationales for Direct Preference Alignment

Mind Control through Causal Inference: Predicting Clean Images from Poisoned Data

LLMs Can Plan Only If We Tell Them

Probing Hidden Knowledge Holes in Unlearned LLMs

LLMs Can Reason Faster Only If We Let Them

Just Enough Shifts: Mitigating Over-Refusal in Aligned Language Models with Targeted Representation Fine-Tuning

AutoScale: Automatic Prediction of Compute-optimal Data Compositions for Training LLMs

AutoScale: Scale-Aware Data Mixing for Pre-Training LLMs

LLM Spark: Critical Thinking Evaluation of Large Language Models

Data-Centric Human Preference Optimization with Rationales

SCOPE: Scalable and Adaptive Evaluation of Misguided Safety Refusal in LLMs

Fast and Noise-Robust Diffusion Solvers for Inverse Problems: A Frequentist Approach

CONCORD: Concept-informed Diffusion for Dataset Distillation