Adaptive Neuron Apoptosis for Accelerating Deep Learning on Large Scale Systems

TL;DR

This paper introduces adaptive neuron apoptosis techniques that accelerate deep learning training by removing redundant neurons during training, achieving significant speedups and parameter reductions while maintaining or improving accuracy.

Contribution

The paper presents novel theoretical and practical methods for adaptive neuron apoptosis, significantly speeding up training and reducing model size without sacrificing accuracy.

Findings

2-3x faster training times on multiple datasets

Up to 30x reduction in parameters, averaging 4-5x

Improved accuracy on Higgs Boson dataset from 0.88/1 to 0.94/1

Abstract

We present novel techniques to accelerate the convergence of Deep Learning algorithms by conducting low overhead removal of redundant neurons -- apoptosis of neurons -- which do not contribute to model learning, during the training phase itself. We provide in-depth theoretical underpinnings of our heuristics (bounding accuracy loss and handling apoptosis of several neuron types), and present the methods to conduct adaptive neuron apoptosis. Specifically, we are able to improve the training time for several datasets by 2-3x, while reducing the number of parameters by up to 30x (4-5x on average) on datasets such as ImageNet classification. For the Higgs Boson dataset, our implementation improves the accuracy (measured by Area Under Curve (AUC)) for classification from 0.88/1 to 0.94/1, while reducing the number of parameters by 3x in comparison to existing literature. The proposed methods achieve a 2.44x speedup in comparison to the default (no apoptosis) algorithm.