Associate Professor, Massachusetts Institute of Technology
3 papers at NeurIPS 2025
We introduce 3BASiL-TM, a highly efficient one-shot post-training method for Sparse plus Low-Rank decomposition of LLMs that reduces the WikiText2 perplexity gap to dense model by over $30\%$ compared to prior methods.
We propose two scalable DP algorithms for high-dimensional sparse variable selection, leveraging modern mixed-integer programming techniques.
