Lazy Evaluation of Convolutional Filters

TL;DR

This paper introduces a lazy evaluation technique for convolutional filters in deep neural networks, reducing computational and memory costs, especially useful for constrained devices, by selectively avoiding certain filter evaluations.

Contribution

It presents a novel method to selectively skip convolutional filter evaluations, balancing accuracy with resource efficiency in deep neural networks.

Findings

Reduces computational load on constrained devices.

Maintains acceptable accuracy levels with selective filter evaluation.

Demonstrates significant memory savings.

Abstract

In this paper we propose a technique which avoids the evaluation of certain convolutional filters in a deep neural network. This allows to trade-off the accuracy of a deep neural network with the computational and memory requirements. This is especially important on a constrained device unable to hold all the weights of the network in memory.