Manifold Identification for Ultimately Communication-Efficient Distributed Optimization

Jul 12, 2020



The expensive inter-machine communication is the bottleneck of distributed optimization. Existing study tackles this problem by shortening the communication rounds, but the reduction of per-round communication cost is not well-studied. This work proposes a progressive manifold identification approach with sound theoretical justifications to greatly reduce both the communication rounds and the bytes communicated per round for partly smooth regularized problems, which include many large-scale machine learning tasks such as the training of ℓ_1- and group-LASSO-regularized models. Our method uses an inexact proximal quasi-Newton method to iteratively identify a sequence of low-dimensional smooth manifolds in which the final solution lies, and restricts the model update within the current manifold to lower significantly the per-round communication cost. After identifying the final manifold within which the problem is smooth, we take superlinear-convergent truncated semismooth Newton steps obtained through preconditioned conjugate gradient to largely reduce the communication rounds. Experiments show that when compared with the state of the art, the communication cost of our method is significantly lower and the running time is up to 10 times faster.


About ICML 2020

The International Conference on Machine Learning (ICML) is the premier gathering of professionals dedicated to the advancement of the branch of artificial intelligence known as machine learning. ICML is globally renowned for presenting and publishing cutting-edge research on all aspects of machine learning used in closely related areas like artificial intelligence, statistics and data science, as well as important application areas such as machine vision, computational biology, speech recognition, and robotics. ICML is one of the fastest growing artificial intelligence conferences in the world. Participants at ICML span a wide range of backgrounds, from academic and industrial researchers, to entrepreneurs and engineers, to graduate students and postdocs.

Store presentation

Should this presentation be stored for 1000 years?

How do we store presentations

Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%


Recommended Videos

Presentations on similar topic, category or speaker