معرفی
Peter Richtarik is a Professor at the King Abdullah University of Science and Technology (KAUST), specializing in Machine Learning, Optimization, and Federated Learning. He actively teaches courses such as Stochastic Gradient Descent Methods and mentors PhD and MS students like Konstantin Burlachenko, Kai Yi, and Lukang Sun. His research spans distributed optimization, parameter-efficient fine-tuning, and theoretical frameworks for non-convex and non-smooth problems.
- Recent work includes 2025 contributions to Bernoulli-LoRA (theoretical PEFT) and Gluon (LMO-based optimizers).
- He co-developed Thanos (block-wise pruning) and BurTorch (CPU-optimized training framework).
His publications focus on communication efficiency (ATA, LoCoDL), differential privacy (DP-RBCD), and stochastic proximal methods. Richtarik received the Charles Broyden Prize for his work on quasi-Newton methods. He frequently presents at workshops like MLSS in Senegal and FLOW seminars.
