Realizing Neural Decoder at the Edge with Ensembled BNN

TL;DR

This paper introduces extreme compression techniques for neural decoders, using binarization and ternarization, and employs ensemble methods to match the performance of real-valued decoders while significantly reducing memory and computation.

Contribution

It proposes a novel ensemble approach to compensate for the limited representation of binarized and ternarized neural decoders, enabling edge deployment with high efficiency.

Findings

Achieves 16-64x savings in memory and computation.

Ensembled weak decoders match real-valued TurboAE performance.

Outperforms quantized neural decoders with similar compression ratios.

Abstract

In this work, we propose extreme compression techniques like binarization, ternarization for Neural Decoders such as TurboAE. These methods reduce memory and computation by a factor of 64 with a performance better than the quantized (with 1-bit or 2-bits) Neural Decoders. However, because of the limited representation capability of the Binary and Ternary networks, the performance is not as good as the real-valued decoder. To fill this gap, we further propose to ensemble 4 such weak performers to deploy in the edge to achieve a performance similar to the real-valued network. These ensemble decoders give 16 and 64 times saving in memory and computation respectively and help to achieve performance similar to real-valued TurboAE.