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


Current browse context:


Change to browse by:


References & Citations

DBLP - CS Bibliography


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

Computer Science > Computation and Language

Title: RankNAS: Efficient Neural Architecture Search by Pairwise Ranking

Abstract: This paper addresses the efficiency challenge of Neural Architecture Search (NAS) by formulating the task as a ranking problem. Previous methods require numerous training examples to estimate the accurate performance of architectures, although the actual goal is to find the distinction between "good" and "bad" candidates. Here we do not resort to performance predictors. Instead, we propose a performance ranking method (RankNAS) via pairwise ranking. It enables efficient architecture search using much fewer training examples. Moreover, we develop an architecture selection method to prune the search space and concentrate on more promising candidates. Extensive experiments on machine translation and language modeling tasks show that RankNAS can design high-performance architectures while being orders of magnitude faster than state-of-the-art NAS systems.
Comments: To appear in EMNLP 2021
Subjects: Computation and Language (cs.CL)
Cite as: arXiv:2109.07383 [cs.CL]
  (or arXiv:2109.07383v2 [cs.CL] for this version)

Submission history

From: Hu Chi [view email]
[v1] Wed, 15 Sep 2021 15:43:08 GMT (179kb,D)
[v2] Fri, 17 Sep 2021 12:32:08 GMT (168kb,D)

Link back to: arXiv, form interface, contact.