Work

Learning Fast Algorithms for Linear Transforms Using Butterfly Factorizations

Year: 2019

Type: article

Abstract: up to 1024. Furthermore, our method can be incorporated as a lightweight replacement of generic matrices in machine learning pipelines to learn efficient and compressible transformations. On a standard task of compressing a single hidden-layer network, our method exceeds the classification accuracy of unconstrained matrices on CIFAR-10 by 3.9 points-the first time a structured approach has done so-with 4× faster inference speed and 40× fewer parameters.

Source: PubMed

DOI: 10.48550/arxiv.1903.05895

Authors: Dao, Tri, Gu, Albert, Eichhorn, Matthew, Rudra, Atri, Ré, Christopher

Institutions: Stanford University, University at Buffalo, State University of New York

Language: English

Topic: Advanced Neural Network Applications

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

Open Access status: green