We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: GroSS: Group-Size Series Decomposition for Grouped Architecture Search

Abstract: We present a novel approach which is able to explore the configuration of grouped convolutions within neural networks. Group-size Series (GroSS) decomposition is a mathematical formulation of tensor factorisation into a series of approximations of increasing rank terms. GroSS allows for dynamic and differentiable selection of factorisation rank, which is analogous to a grouped convolution. Therefore, to the best of our knowledge, GroSS is the first method to enable simultaneously train differing numbers of groups within a single layer, as well as all possible combinations between layers. In doing so, GroSS is able to train an entire grouped convolution architecture search-space concurrently. We demonstrate this through architecture searches with performance objectives and evaluate its performance against conventional Block Term Decomposition. GroSS enables more effective and efficient search for grouped convolutional architectures.
Subjects: Machine Learning (cs.LG); Machine Learning (stat.ML)
Cite as: arXiv:1912.00673 [cs.LG]
  (or arXiv:1912.00673v2 [cs.LG] for this version)

Submission history

From: Henry Howard-Jenkins [view email]
[v1] Mon, 2 Dec 2019 10:32:50 GMT (119kb,D)
[v2] Mon, 23 Mar 2020 12:26:25 GMT (4399kb,D)
[v3] Thu, 16 Jul 2020 16:28:12 GMT (4659kb,D)

Link back to: arXiv, form interface, contact.