Newsroom
Optimization
-
FDS Colloquium: Lorenzo Orecchia (Chicago), “Variational Characterizations of First-Order Algorithms via Self-Duality”
Talk summary: First-order methods for convex optimization play an important role in the efficient deployment of machine learning algorithms. While a large number of such methods exist, each tuned to the specific properties of the problem under consideration, it is not always clear how to generalize their approach to a new setting or to a…
-
FDS Colloquium: George Lan (Georgia Tech), “Algorithmic Foundations of Risk-averse Optimization for Trustworthy AI”
Talk summary: Over the past two decades, stochastic optimization has made remarkable strides, driving its widespread adoption in machine learning (ML) and artificial intelligence (AI). However, most existing models prioritize minimizing expected loss, often leaving AI-driven decisions vulnerable to costly or catastrophic failures and raising concerns about their trustworthiness in high-stakes applications. Risk-averse optimization provides a principled…
-
FDS Colloquium: Pravesh Kothari (Princeton), “The surprising reach of spectral algorithms for smoothed k-SAT”
Abstract: Semirandom input models are hybrids of the classical worst-case and average-case models in algorithm design. They were introduced in the 1990s to inspire “robust heuristics” that, on the one hand, escape worst-case hardness results, while, on the other, avoid “overfitting” to a specific distribution of input instances. Over the past five years, rapid progress…
-
FDS Colloquium: Elliot Paquette (McGill), “High-dimensional Optimization with Applications to Compute-Optimal Neural Scaling Laws”
Abstract: Given the massive scale of modern ML models, we now only get a single shot to train them effectively. This restricts our ability to test multiple architectures and hyper-parameter configurations. Instead, we need to understand how these models scale, allowing us to experiment with smaller problems and then apply those insights to larger-scale models. In this talk,…
-
FDS x Applied Physics Colloquium: Grant Rotskoff (Stanford), “Efficient variational inference with generative models”
Abstract: Neural networks continue to surprise us with their remarkable capabilities for high-dimensional function approximation. Applications of machine learning now pervade essentially every scientific discipline, but predictive models to describe the optimization dynamics, inference properties, and flexibility of modern neural networks remain limited. In this talk, I will introduce several approaches to both analyzing and…
-
FDS Colloquium: Brice Huang (MIT), “Algorithmic thresholds in random optimization problems”
Abstract: Optimizing high-dimensional functions generated from random data is a central problem in modern statistics and machine learning. As these objectives are highly non-convex, the maximum value reachable by efficient algorithms is usually smaller than the maximum value that exists, and characterizing the fundamental computational limits of these problems is a difficult challenge. In this…