Beyond Black-Box Scaling: Interpretability Algorithms and GPUs.

Abstract

As machine learning becomes more widespread, it is increasingly used to support high-stakes decisions in domains such as healthcare and scientific discovery. However, as models become larger and more complex, we often lose control: black-box models can be difficult to troubleshoot, constrain with domain requirements, and trust in deployment. How can we extract actionable insights beyond prediction, and how can we better leverage today's growing compute to do so?

 

This talk explores an alternative route: scaling optimization while shrinking model complexity, especially for problems with discrete structure --- often yielding 10--100x reductions in training time relative to traditional approaches. First, I will show how scalable discrete optimization can learn interpretable medical scoring systems from modern real-world datasets that achieve black-box-level accuracy while remaining deployable as simple, auditable, and interactive decision tools. Second, I will describe how GPU-compatible methods make it practical to obtain certifiably optimal solutions for highly nonconvex, combinatorial problems, enabling new capabilities such as discovering governing differential equations from data in physical science.

 

 

Speaker: Dr. Jiachang LIU
Date: 18 Mar 2026 (Monday)
Time: 9:30am – 10:30am
Zoom: Link
Poster: Click here