Pruning In Time (PIT): A Lightweight Network Architecture Optimizer for Temporal Convolutional Networks

Temporal Convolutional Networks (TCNs) are promising Deep Learning models for time-series processing tasks. One key feature of TCNs is time-dilated convolution, whose optimization requires extensive experimentation. We propose an automatic dilation optimizer, which tackles the problem as a weight pr...

Full description

Saved in:

Bibliographic Details
Published in:	2021 58th ACM/IEEE Design Automation Conference (DAC) pp. 1015 - 1020
Main Authors:	Risso, Matteo, Burrello, Alessio, Pagliari, Daniele Jahier, Conti, Francesco, Lamberti, Lorenzo, Macii, Enrico, Benini, Luca, Poncino, Massimo
Format:	Conference Proceeding
Language:	English
Published:	IEEE 05.12.2021
Subjects:	Computer architecture Deep learning Design automation Edge Computing Network architecture Neural Architecture Search Neural networks Temporal Convolutional Networks Training
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	Temporal Convolutional Networks (TCNs) are promising Deep Learning models for time-series processing tasks. One key feature of TCNs is time-dilated convolution, whose optimization requires extensive experimentation. We propose an automatic dilation optimizer, which tackles the problem as a weight pruning on the time-axis, and learns dilation factors together with weights, in a single training. Our method reduces the model size and inference latency on a real SoC hardware target by up to 7.4× and 3×, respectively with no accuracy drop compared to a network without dilation. It also yields a rich set of Pareto-optimal TCNs starting from a single model, outperforming hand-designed solutions in both size and accuracy.
DOI:	10.1109/DAC18074.2021.9586187