KAKURENBO: Adaptively Hiding Samples in Deep Neural Network Training

TL;DR

This paper introduces KAKURENBO, a method that adaptively hides less important samples during deep neural network training to improve efficiency with minimal accuracy loss, reducing training time by up to 22%.

Contribution

It presents a novel adaptive sample hiding technique based on loss and confidence, improving training efficiency without significantly degrading accuracy.

Findings

Reduces training time by up to 22%.

Maintains accuracy within 0.4% of baseline.

Effective on large-scale image datasets.

Abstract

This paper proposes a method for hiding the least-important samples during the training of deep neural networks to increase efficiency, i.e., to reduce the cost of training. Using information about the loss and prediction confidence during training, we adaptively find samples to exclude in a given epoch based on their contribution to the overall learning process, without significantly degrading accuracy. We explore the converge properties when accounting for the reduction in the number of SGD updates. Empirical results on various large-scale datasets and models used directly in image classification and segmentation show that while the with-replacement importance sampling algorithm performs poorly on large datasets, our method can reduce total training time by up to 22% impacting accuracy only by 0.4% compared to the baseline. Code available at https://github.com/TruongThaoNguyen/kakurenbo