Progressive Data Dropout: An Embarrassingly Simple Approach to Faster Training

TL;DR

Progressive Data Dropout is a simple, model-agnostic training method that significantly reduces training time by dropping data progressively, while maintaining or improving accuracy, offering a practical alternative to traditional uniform sampling.

Contribution

The paper introduces Progressive Data Dropout, a novel training paradigm that decreases effective epochs drastically without sacrificing accuracy, requiring no changes to existing models or training pipelines.

Findings

Reduces effective epochs to 12.4% of baseline

Maintains or improves accuracy by up to 4.82%

Easy to implement and widely applicable

Abstract

The success of the machine learning field has reliably depended on training on large datasets. While effective, this trend comes at an extraordinary cost. This is due to two deeply intertwined factors: the size of models and the size of datasets. While promising research efforts focus on reducing the size of models, the other half of the equation remains fairly mysterious. Indeed, it is surprising that the standard approach to training remains to iterate over and over, uniformly sampling the training dataset. In this paper we explore a series of alternative training paradigms that leverage insights from hard-data-mining and dropout, simple enough to implement and use that can become the new training standard. The proposed Progressive Data Dropout reduces the number of effective epochs to as little as 12.4% of the baseline. This savings actually do not come at any cost for accuracy. Surprisingly, the proposed method improves accuracy by up to 4.82%. Our approach requires no changes to model architecture or optimizer, and can be applied across standard training pipelines, thus posing an excellent opportunity for wide adoption. Code can be found here: https://github.com/bazyagami/LearningWithRevision