| Description |
- Sequential decision-making is at the heart of personalized medicine and adaptive clinical trials, where treatments must be adjusted as data accumulate. While multi-armed bandit methods are well suited to these problems, real-world data introduce complications like delayed feedback, patient characteristics, and related treatment options.
In this talk, Song will present a new framework for batched bandit problems that accounts for these challenges. The approach uses a flexible yet interpretable model to learn how treatments are related and introduces a new algorithm that efficiently narrows down the best options over time. Through simulations and real-world examples, Song will show how this method improves decision-making accuracy and outperforms existing approaches in complex clinical settings.
|