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: Dynamic Precision Analog Computing for Neural Networks

Abstract: Analog electronic and optical computing exhibit tremendous advantages over digital computing for accelerating deep learning when operations are executed at low precision. In this work, we derive a relationship between analog precision, which is limited by noise, and digital bit precision. We propose extending analog computing architectures to support varying levels of precision by repeating operations and averaging the result, decreasing the impact of noise. Such architectures enable programmable tradeoffs between precision and other desirable performance metrics such as energy efficiency or throughput. To utilize dynamic precision, we propose a method for learning the precision of each layer of a pre-trained model without retraining network weights. We evaluate this method on analog architectures subject to a variety of noise sources such as shot noise, thermal noise, and weight noise and find that employing dynamic precision reduces energy consumption by up to 89% for computer vision models such as Resnet50 and by 24% for natural language processing models such as BERT. In one example, we apply dynamic precision to a shot-noise limited homodyne optical neural network and simulate inference at an optical energy consumption of 2.7 aJ/MAC for Resnet50 and 1.6 aJ/MAC for BERT with <2% accuracy degradation.
Subjects: Machine Learning (cs.LG); Emerging Technologies (cs.ET)
Cite as: arXiv:2102.06365 [cs.LG]
  (or arXiv:2102.06365v1 [cs.LG] for this version)

Submission history

From: Sahaj Garg [view email]
[v1] Fri, 12 Feb 2021 06:56:56 GMT (2022kb,D)

Link back to: arXiv, form interface, contact.