Optimizing Temporal Convolutional Networks for Ultra-Low-Power Edge-Based Time Series Analytics
TimeTuesday, December 7th6:00pm - 7:00pm PST
LocationLevel 2 - Lobby
Event Type
Networking Reception
Work-in-Progress Poster
Virtual Programs
Presented In-Person
DescriptionTemporal Convolutional Networks (TCNs) are emerging light-weight DL models for Time Series analysis. We introduce a toolkit and a library of highly optimized kernels to efficiently map TCNs on Parallel Ultra-Low Power microcontrollers. Our approach minimizes latency and energy consumption by selecting among alternative implementations of the causal and dilated 1D-convolution operations at the core of TCNs. We benchmark our approach on a commercial PULP device, achieving up to 103x lower latency and 20.3x lower energy than the Cube-AI toolkit and from 3x to 26.6x better performance compared to commercial closed-source and academic open-source approaches on the same hardware target.