About Me
I am a Senior Machine Learning Engineer at Uber and an Adjunct Assistant Professor in the CSE department at The Ohio State University. My research focuses on LLM alignment and post-training, mechanistic interpretability, and efficient machine learning, including model compression and knowledge distillation.
Previously, I was a tenure-track Assistant Professor at Ohio State (2023–2025), a Research Scientist at Yahoo! Research (2022–2024), and an Assistant Professor of Computer Science at the University of Delaware (2020–2022). I was a postdoctoral researcher in EECS at UC Berkeley, and received my Ph.D. in EECS and M.Sc. in Mathematics from the University of Michigan, and my B.Sc. and M.Sc. in Electrical Engineering from Sharif University of Technology.
I have had the privilege of working with and advising talented students, including Zhiqun Zuo, Zhongteng Cai, Ding Zhu, Vishnu Chhabra, and Suchit Gupte.
Selected Publications
D. Zhu, M. Khalili, “WAQS: Weight Absorption for Quadratic Probing and Affine Steering,” under review, 2026. [code]
D. Zhu, X. Wei, T. Xie, Z. Zhu, X. Zhang, M. Khalili, “PORT: Preference Optimization via Robust Token-Level Reweighting,” The Conference on Neural Information Processing Systems (NeurIPS), 2026. [code]
S. Gupte, X. Zhang, M. Khalili, “When Pruning Meets Interpretability: Preserving Sparse Autoencoder Robustness in LLMs,” The Conference on Language Modeling (COLM), 2026. [paper] [code]
X. Zhu, M. Khalili, Z. Zhu, “AbsTopK: Rethinking Sparse Autoencoders for Bidirectional Features,” The International Conference on Learning Representations (ICLR), 2026. [paper] [code]
X. Zhu, J. Jiang, M. Khalili, Z. Zhu, “From Emergence to Control: Probing and Modulating Self-Reflection in Language Models,” Transactions on Machine Learning Research (TMLR), 2026. [paper] [code]
D. Zhu, Z. Zuo, M. Khalili, “An Efficient Training Algorithm for Models with Block-wise Sparsity,” Transactions on Machine Learning Research (TMLR), 2025. [paper] [code]
See all publications on the Publications page.