Hardware-Aware Training and Inference for Large-Scale AI

Animashree Anandkumar

Conference on Machine Learning and Systems 2025 · Day 3 · Invited Talk

In an era where large-scale AI models are continually pushing the boundaries of computational resources, Professor Animashree Anandkumar's talk at MLSys 2025 presents a compelling vision for the future of machine learning systems: **co-designing algorithms and hardware**. Moving beyond the traditional paradigm of optimizing systems for fixed machine learning architectures, Anandkumar advocates for a fundamental reimagining of ML algorithms themselves to better suit the constraints and opportunities presented by modern hardware. This approach is not merely about incremental improvements through pruning or quantization but about architecting algorithms from first principles with hardware efficiency in mind.

AI review

Anandkumar covers genuinely important territory — hardware-aware co-design, LNS training, Galore, Neural Operators — and the results cited are real and significant. But this writeup (and presumably the talk itself) reads like a well-organized survey of her group's portfolio rather than a deep engineering walkthrough of any single system. The headline numbers are impressive, but I can't reproduce any of this from what's here, and the talk doesn't seem to have committed to showing how any one technique actually works at the implementation level.