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: TENT: Efficient Quantization of Neural Networks on the tiny Edge with Tapered FixEd PoiNT

Abstract: In this research, we propose a new low-precision framework, TENT, to leverage the benefits of a tapered fixed-point numerical format in TinyML models. We introduce a tapered fixed-point quantization algorithm that matches the numerical format's dynamic range and distribution to that of the deep neural network model's parameter distribution at each layer. An accelerator architecture for the tapered fixed-point with TENT framework is proposed. Results show that the accuracy on classification tasks improves up to ~31 % with an energy overhead of ~17-30 % as compared to fixed-point, for ConvNet and ResNet-18 models.
Comments: poster presented at the first tinyML Research Symposium, March 26, 2021
Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
Cite as: arXiv:2104.02233 [cs.LG]
  (or arXiv:2104.02233v1 [cs.LG] for this version)

Submission history

From: Seyed Hamed Fatemi Langroudi [view email]
[v1] Tue, 6 Apr 2021 01:54:32 GMT (3135kb,D)

Link back to: arXiv, form interface, contact.