Machine Learning Seminar: Dr. Dravyansh Sharma, "New Frontiers for Data-driven Algorithm Design."
- Sponsor
- Siebel School of Computing and Data Science
- Speaker
- Dr. Dravyansh Sharma
- Contact
- Nishant Jain
- nj27@illinois.edu
- Originating Calendar
- Siebel School Speakers Calendar
- Abstract: Modern algorithms, especially in machine learning, often come with a rich set of design choices—learning rates, regularization parameters, clustering or branching heuristics, and other hyperparameters—that can be crucial to an algorithm’s practical effectiveness. Data-driven algorithm design (D-DAD) provides a principled framework for using past problem instances to automatically tune these choices while providing rigorous performance guarantees.
In this talk, Dravyansh will describe several recent directions that broaden the scope of this framework beyond the discrete and combinatorial algorithm families that motivated much of the early work in the area. He will discuss how to analyze continuous and iterative algorithms, where performance as a function of the parameters can exhibit considerably richer structure. Examples include provably tuning regularization in linear regression and learning rates for gradient-based optimization. He will also describe some new general techniques for controlling the statistical complexity of parameterized algorithms. He will conclude with a discussion of future directions and open problems, including better online algorithm configuration, stronger computational guarantees, and more powerful generalization guarantees.
The talk connects to themes from recent tutorials at COLT 2026 and NeurIPS 2025.
Bio: Dravyansh Sharma is a Postdoctoral Researcher at TTIC working with Avrim Blum. He received his Ph.D. from CMU, advised by Nina Balcan. His research lies at the intersection of machine learning, algorithms, and optimization, with a particular focus on data-driven algorithm design, the foundations and modern challenges of reliable AI, and learning with agents. His work has received the UAI 2024 Outstanding Student Paper Award and he has recently presented tutorials at COLT, NeurIPS, UAI, and AutoML.