IAS Seminar #09/2026 Sequential Decision-Making: from Obs..Optimal Decisions – Nicolò Cesa-Bianchi

Sequential Decision-Making: from Observations to Optimal Decisions Nicolò Cesa-Bianchi While machine learning is classically linked to statistical analysis, the paradigm of online learning provides an alternative approach where models are sequentially trained on data streams and fast learning rates are obtained without relying on statistical assumptions. Besides creating learning algorithms that are robust against arbitrary data sources, online learning is a natural framework for the study of settings where the actions of the decision maker actively shape the feedback received from the environment. This adds a new dimension to the design learning algorithms; namely, identifying the best possible learning curve for a given feedback structure. In this talk, we will describe the core algorithmic toolkit of modern online learning and establish rigorous mathematical guarantees for their behavior. Finally, we will explore how these sequential tools bridge the gap to broader domains, expanding our ability to move from observations to optimal decisions.