Class-Level Logit Perturbation

TL;DR

This paper explores class-level logit perturbation in neural networks, providing a theoretical foundation, new learning methodologies, and demonstrating competitive results on image classification benchmarks, including long-tail datasets.

Contribution

It introduces explicit logit perturbation methods for classification, unifies existing data augmentation and loss variation perspectives, and offers a plug-in approach adaptable to various algorithms.

Findings

Effective in improving model robustness and generalization.

Competitive performance on standard and long-tail datasets.

Applicable as a plug-in to existing classification methods.

Abstract

Features, logits, and labels are the three primary data when a sample passes through a deep neural network. Feature perturbation and label perturbation receive increasing attention in recent years. They have been proven to be useful in various deep learning approaches. For example, (adversarial) feature perturbation can improve the robustness or even generalization capability of learned models. However, limited studies have explicitly explored for the perturbation of logit vectors. This work discusses several existing methods related to class-level logit perturbation. A unified viewpoint between positive/negative data augmentation and loss variations incurred by logit perturbation is established. A theoretical analysis is provided to illuminate why class-level logit perturbation is useful. Accordingly, new methodologies are proposed to explicitly learn to perturb logits for both single-label and multi-label classification tasks. Extensive experiments on benchmark image classification data sets and their long-tail versions indicated the competitive performance of our learning method. As it only perturbs on logit, it can be used as a plug-in to fuse with any existing classification algorithms. All the codes are available at https://github.com/limengyang1992/lpl.