Samet Oymak

Associate Professor
EECS Department, University of Michigan

Adjunct Professor
ECE Department, UC Riverside

Email: lastname (at) umich (dot) edu

PhD in Electrical Engineering, California Institute of Technology, 2015

SOTA Lab advances Statistics & Optimization for Trustworthy AI.

UPDATE: Our group has moved to the University of Michigan: SOTA Lab website

Research focus: We develop theoretically-grounded and empirically-impactful AI/ML methods.

Current research topics:

mathematical foundations of transformers and attention
trustworthy and efficient language models, LLM systems
reinforcement learning, control, LLMs as interactive agents
core optimization and statistical learning theory

Research sponsors

News (last update 09/2023)

09/2023 Our new work demystifies the optimization foundations of transformers via SVM connection: Link to Twitter thread
09/2023 New paper on attention mechanism will appear at NeurIPS 2023 as spotlight.
09/2023 New paper demystifies compositional learning abilities of chain-of-thought prompting, NeurIPS 2023.
07/2023 New paper develops the first federated multi-Sequence stochastic approximation methods.
07/2023 Necmiye and I received a Propelling Original Data Science grant from MIDAS.
07/2023 Two transformer theory papers will appear at ICML23: attention and prompt-tuning, in-context learning
04/2023 “Learning on Manifolds: Universal Approximation Properties of Neural ODEs” is accepted as oral presentation at L4DC.
04/2023 “On the Role of Attention in Prompt-tuning” appeared at ICLR Foundation Models workshop.
02/2023 Our new paper interprets transformers as learning algorithms and characterizes their generalization and stability.
02/2023 Two papers are accepted to AAAI23. Congrats to collaborators and Yingcong!
- Y. Li and S. Oymak, “Provable Pathways: Learning Multiple Tasks over Multiple Paths”.
- Y. Qin, Y. Li, F. Pasqualetti, M. Fazel, S. Oymak, “Stochastic Contextual Bandits with Long Horizon Rewards”.
11/2022 I am giving seminars at Google NY, KTH, and UCR on our recent results.
09/2022 We welcome Emrullah to our team!
07/2022 Nael, Khaled and I received an NSF medium award on approximate computing for ML security
07/2022 Congrats to Mingchen on receiving DYP fellowship from UCR graduate division!
07/2022 Two papers are accepted to IEEE CDC 2022 and Open Journal of Control Systems
05/2022 FedNest is accepted to ICML’22 as oral presentation
05/2022 Our paper Non-Stationary Representation Learning in Sequential Linear Bandits is accepted to IEEE Open Journal of Control Systems
04/2022 Received a Google Research Scholar award!
03/2022 Karthik will be joining our lab and MURI team as a postdoctoral scholar!
03/2022 New paper on Provable and Efficient Continual Representation Learning.
02/2022 Three papers accepted to ACC 2022.
02/2022 New paper on “FedNest: Federated Bilevel, Minimax, and Compositional Optimization”.
01/2022 Our paper on “Non-asymptotic and accurate learning of nonlinear dynamical systems” is accepted to JMLR.
01/2022 ML workshop at RoseHack 2021 on “Challenges & Opportunities in Machine Learning”.