vix.ing · top · new · best · stats · spec

On the Computational Power of Online Gradient Descent

2018/07/03 by Vaggos Chatziafratis, Tim Roughgarden, Chatziafratis, Vaggos +3
Computer Science · Decision Sciences · #Advanced Bandit Algorithms Research #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Machine Learning and Algorithms #Stochastic Gradient Optimization Techniques

paper · pdf · doi:10.48550/arxiv.1807.01280

openalex publication_date 2018/07/03 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

We prove that the evolution of weight vectors in online gradient descent can encode arbitrary polynomial-space computations, even in very simple learning settings. Our results imply that, under weak complexity-theoretic assumptions, it is impossible to reason efficiently about the fine-grained behavior of online gradient descent.

Citations

Related