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.
Titolo: | Pruning in Time (PIT): A Lightweight Network Architecture Optimizer for Temporal Convolutional Networks | |
Autore/i: | Risso M.; Burrello A.; Pagliari D. J.; Conti F.; Lamberti L.; MacIi E.; Benini L.; Poncino M. | |
Autore/i Unibo: | ||
Anno: | 2021 | |
Titolo del libro: | Proceedings - Design Automation Conference | |
Pagina iniziale: | 1015 | |
Pagina finale: | 1020 | |
Digital Object Identifier (DOI): | http://dx.doi.org/10.1109/DAC18074.2021.9586187 | |
Abstract: | 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. | |
Data stato definitivo: | 23-gen-2022 | |
Appare nelle tipologie: | 4.01 Contributo in Atti di convegno |