Sequenced-Replacement Sampling for Deep Learning

TL;DR

This paper introduces sequenced-replacement sampling (SRS), a novel training method for deep neural networks that enhances performance and reduces overfitting, especially on datasets with few images per class like CIFAR-100.

Contribution

The paper presents SRS, a new sampling technique that improves deep network training by effectively augmenting mini-batches and reducing overfitting on small datasets.

Findings

Significantly improves CIFAR-100 accuracy over state-of-the-art.

Training deeper networks with SRS reduces overfitting.

Achieves as low as 10.10% error rate.

Abstract

We propose sequenced-replacement sampling (SRS) for training deep neural networks. The basic idea is to assign a fixed sequence index to each sample in the dataset. Once a mini-batch is randomly drawn in each training iteration, we refill the original dataset by successively adding samples according to their sequence index. Thus we carry out replacement sampling but in a batched and sequenced way. In a sense, SRS could be viewed as a way of performing "mini-batch augmentation". It is particularly useful for a task where we have a relatively small images-per-class such as CIFAR-100. Together with a longer period of initial large learning rate, it significantly improves the classification accuracy in CIFAR-100 over the current state-of-the-art results. Our experiments indicate that training deeper networks with SRS is less prone to over-fitting. In the best case, we achieve an error rate as low as 10.10%.