Willie Neiswanger

Assistant Professor@University of Southern California·美国·OpenReview

研究方向

probabilistic machine learning · approximate inference · hyperparameter optimization · neural architectures · experimental design · active learning · bayesian optimization · ai for science · ai for engineering · generative models · large language models · LLMs

Willie Neiswanger

Hubble: a Model Suite to Advance the Study of LLM Memorization

Zebra-CoT: A Dataset for Interleaved Vision-Language Reasoning

Tina: Tiny Reasoning Models via LoRA

Demystifying and Enhancing the Efficiency of Large Language Model Based Search Agents

Auditing Black-Box LLM APIs with a Rank-Based Uniformity Test

Euclid: Lessons for Geometric Low-level Visual Perception in Multimodal LLMs

Resa: Efficient Reasoning Models via SAEs

Informative Posterior Ensembles for Sequential Simulation-Based Inference

Textual Steering Vectors Can Improve Visual Understanding in Multimodal Large Language Models

LLM Unlearning Without an Expert Curated Dataset

DeLLMa: Decision Making Under Uncertainty with Large Language Models

LiveBench: A Challenging, Contamination-Limited LLM Benchmark

What is Your Data Worth to GPT? LLM-Scale Data Valuation with Influence Functions

Sample Efficient Preference Alignment in LLMs via Active Exploration

Sharpe Ratio-Guided Active Learning for Preference Optimization in RLHF

Large-Scale Training Data Attribution with Efficient Influence Functions

Euclid: Supercharging Multimodal LLMs with Synthetic High-Fidelity Visual Descriptions

Nonmyopic Bayesian Optimization in Dynamic Cost Settings

Mechanistic Insights: Circuit Transformations Across Input and Fine-Tuning Landscapes