Label-similarity Curriculum Learning

TL;DR

This paper introduces a curriculum learning method that gradually shifts label representations towards one-hot encoding based on class similarity, improving neural network training for image classification.

Contribution

It proposes a novel label-similarity curriculum learning approach that adapts loss functions by using class similarity, including a method to automatically compute similarities from word embeddings.

Findings

LCL improves classification accuracy across multiple datasets.

LCL outperforms standard training methods.

The approach is effective with various deep learning architectures.

Abstract

Curriculum learning can improve neural network training by guiding the optimization to desirable optima. We propose a novel curriculum learning approach for image classification that adapts the loss function by changing the label representation. The idea is to use a probability distribution over classes as target label, where the class probabilities reflect the similarity to the true class. Gradually, this label representation is shifted towards the standard one-hot-encoding. That is, in the beginning minor mistakes are corrected less than large mistakes, resembling a teaching process in which broad concepts are explained first before subtle differences are taught. The class similarity can be based on prior knowledge. For the special case of the labels being natural words, we propose a generic way to automatically compute the similarities. The natural words are embedded into Euclidean space using a standard word embedding. The probability of each class is then a function of the cosine similarity between the vector representations of the class and the true label. The proposed label-similarity curriculum learning (LCL) approach was empirically evaluated using several popular deep learning architectures for image classification tasks applied to five datasets including ImageNet, CIFAR100, and AWA2. In all scenarios, LCL was able to improve the classification accuracy on the test data compared to standard training.