Low-resolution neural networks represent both weights and activations with few bits, drastically reducing the multiplication complexity. Nonetheless, these products are accumulated using high-resolution (typically 32-bit) additions, an operation that dominates the arithmetic complexity of inference when using extreme quantization (e.g., binary weights). To further optimize inference, we propose a method that adapts neural networks to use low-resolution (8-bit) additions in the accumulators, achieving classification accuracy comparable to their 32-bit counterparts. We achieve resilience to low-resolution accumulation by inserting a cyclic activation layer, as well as an overflow penalty regularizer. We demonstrate the efficacy of our approach on both software and hardware platforms.
@article{arxiv.2007.13242,
title = {WrapNet: Neural Net Inference with Ultra-Low-Resolution Arithmetic},
author = {Renkun Ni and Hong-min Chu and Oscar Castañeda and Ping-yeh Chiang and Christoph Studer and Tom Goldstein},
journal= {arXiv preprint arXiv:2007.13242},
year = {2020}
}