SemifreddoNets: Partially Frozen Neural Networks for Efficient Computer Vision Systems

TL;DR

SemifreddoNets are partially frozen neural networks designed for efficient hardware implementation, significantly reducing silicon area and power consumption while maintaining flexibility through configurable weights and repeatable blocks.

Contribution

Introduction of SemifreddoNets, a novel partially frozen neural network architecture optimized for hardware efficiency with configurable weights and scalable complexity.

Findings

Up to tenfold reduction in silicon area.

Significant power savings compared to general-purpose accelerators.

Maintains flexibility with configurable weights and repeatable blocks.

Abstract

We propose a system comprised of fixed-topology neural networks having partially frozen weights, named SemifreddoNets. SemifreddoNets work as fully-pipelined hardware blocks that are optimized to have an efficient hardware implementation. Those blocks freeze a certain portion of the parameters at every layer and replace the corresponding multipliers with fixed scalers. Fixing the weights reduces the silicon area, logic delay, and memory requirements, leading to significant savings in cost and power consumption. Unlike traditional layer-wise freezing approaches, SemifreddoNets make a profitable trade between the cost and flexibility by having some of the weights configurable at different scales and levels of abstraction in the model. Although fixing the topology and some of the weights somewhat limits the flexibility, we argue that the efficiency benefits of this strategy outweigh the advantages of a fully configurable model for many use cases. Furthermore, our system uses repeatable blocks, therefore it has the flexibility to adjust model complexity without requiring any hardware change. The hardware implementation of SemifreddoNets provides up to an order of magnitude reduction in silicon area and power consumption as compared to their equivalent implementation on a general-purpose accelerator.