I am Shervin Khalafi, a Fourth-year PhD student of Electrical & Systems Engineeering at the University of Pennsylvania where I am advised by Prof. Alejandro Ribeiro. My research inteterests include Generative Models, Optimization Theory, and graph-based Machine Learning.

Currently, I am developing constrained optimization frameworks for generative model training and inference under requirements, with a particular focus on diffusion and flow models. I am also studying generative diffusion models for Graphs.

News

1. (05/2026) Our paper “Unlearning in Diffusion Models: A Unified Framework with KL Divergence and Likelihood Constraints” was accepted in ICML 2026!
2. (09/2025) Our paper “Composition and Alignment of Diffusion Models using Constrained Learning” was accepted in NeurIPS 2025!
3. (09/2024) Our paper “Constrained Diffusion Models via Dual Training” was accepted in NeurIPS 2024.
4. (05/2024) Our paper “Neural Tangent Kernels Motivate Cross-Covariance Graphs in Neural Networks” was accepted in ICML 2024.