ATCN
Resource-Efficient Processing of Time Series on Edge
ATCN (Agile Temporal Convolutional Network) enables fast time-series prediction and accurate classification in resource-constrained embedded systems.
Key Features
- Family of compact networks with formalized hyperparameters for application-specific architecture tuning
- Deployed on Cortex-M7 microcontroller consuming only 49% of 320KB RAM and 15% of 1MB flash
- Validated across 70 UCR 2018 time-series benchmarks
- Data augmentation techniques: jittering, magnitude warping, window warping, scaling
Publication
M. Baharani and H. Tabkhi, “ATCN: Resource-Efficient Processing of Time Series on Edge,” ACM Transactions on Embedded Computing Systems, vol. 21, no. 5, pp. 1-19, 2022.
| Paper | Code |