Ben Recht • 12/10/2025

Benchmark Studies

This article, based on a final lecture from a 2025 graduate machine learning class, explores the tension between technical innovation and technical debt. The author argues that machine learning theory is less mathematical than other fields, as it often relies on the axiom of generalization rather than provable theorems. It critiques the pursuit of optimal constants for theoretical bounds that give bad practical advice, emphasizing that real progress comes from engineering best practices and social analysis, not just functional analysis.

0 comments

#Machine Learning #technical debt #optimization