Algorithms with Predictions: What Comes Next?
Saturday, August 15, 9:00–10:00
Algorithms with predictions have emerged as one of the most influential developments in algorithms research over the past decade. By augmenting classical algorithmic decision-making with machine-learned forecasts, this paradigm has led to significant advances across a wide range of optimization and online decision problems. As the field matures, however, it is natural to ask: what are the next fundamental questions and challenges that will shape its future?
In this talk, I will discuss two research directions that I find particularly exciting. The first part will explore the search for broad algorithmic principles and design frameworks that transcend individual problems, moving beyond techniques tailored to specific applications toward a more unified theory of algorithms with predictions. The second part will focus on robustness: how predictions fail, the different ways in which they can be unreliable, and how algorithms can be designed to recover gracefully from such errors while retaining strong performance guarantees.
The talk will provide a self-contained overview of these directions and highlight opportunities for future research in these areas.
The Price of Smoothness in Learning-Augmented First-Price Auctions
Saturday, August 15, 14:00–15:00
In first-price auctions, bidders must decide how much to shade their bid below their value—bid too low and you lose, bid too high and you leave money on the table. A natural idea is to use predictions of the competition, but first-price auctions have an awkward feature: get the prediction wrong by even a little and you can lose everything, making it hard for prediction-based strategies to fail gracefully.
In this talk, I will discuss what performance guarantees are actually achievable once we insist on this kind of graceful degradation. I will show that in a single auction, smoothness comes at a real cost to performance under good predictions—but that repeated play offers a way out, via a simple algorithm that adaptively learns how much to trust its predictions.