Complex Deep Neural Networks such as Capsule Networks (CapsNets) exhibit high learning capabilities at the cost of compute-intensive operations. To enable their deployment on edge devices, we propose to leverage approximate computing for designing approximate variants of the complex operations like softmax and squash. In our experiments, we evaluate tradeoffs between area, power consumption, and critical path delay of the designs implemented with the ASIC design flow, and the accuracy of the quantized CapsNets, compared to the exact functions.
@article{arxiv.2206.10200,
title = {Enabling Capsule Networks at the Edge through Approximate Softmax and Squash Operations},
author = {Alberto Marchisio and Beatrice Bussolino and Edoardo Salvati and Maurizio Martina and Guido Masera and Muhammad Shafique},
journal= {arXiv preprint arXiv:2206.10200},
year = {2022}
}
Comments
To appear at the ACM/IEEE International Symposium on Low Power Electronics and Design (ISLPED), August 2022, Boston, MA, USA