Sujay Sanghavi

Associate Professor@University of Texas, Austin·美国·OpenReview

研究方向

optimization · robustness · sparsity · high dimensional statistics · active learning · theory · language models · transformer architecture · parameter efficient fine tuning · distillation

7.2

Upweighting Easy Samples in Fine-Tuning Mitigates Forgetting

ICML 2025Spotlight

通讯

7.0

Retraining with Predicted Hard Labels Provably Increases Model Accuracy

ICML 2025Poster

6.3

Infilling Score: A Pretraining Data Detection Algorithm for Large Language Models

合作者 (20)

Sujay Sanghavi

Upweighting Easy Samples in Fine-Tuning Mitigates Forgetting

Retraining with Predicted Hard Labels Provably Increases Model Accuracy

Infilling Score: A Pretraining Data Detection Algorithm for Large Language Models

Learning Mixtures of Experts with EM: A Mirror Descent Perspective

Enhancing Language Model Agents using Diversity of Thoughts

Geometric Median (GM) Matching for Robust k-Subset Selection from Noisy Data

RARe: Retrieval Augmented Retrieval with In-Context Examples

Inheritune: Training Smaller Yet More Attentive Language Models

Geometric Median (GM) Matching for Robust Data Pruning

Retraining with Predicted Hard Labels Provably Increases Model Accuracy