Interpret Your Decision: Logical Reasoning Regularization for Generalization in Visual Classification

TL;DR

This paper introduces L-Reg, a logical regularization method that improves the generalization of vision models by reducing complexity and enhancing interpretability, especially in unseen domains and novel categories.

Contribution

The paper proposes L-Reg, a novel logical regularization technique that links logical reasoning with deep learning, improving interpretability and generalization in visual classification tasks.

Findings

L-Reg reduces model complexity in feature distribution and classifier weights.

L-Reg enhances interpretability by extracting salient features.

L-Reg improves generalization in multi-domain and category discovery scenarios.

Abstract

Vision models excel in image classification but struggle to generalize to unseen data, such as classifying images from unseen domains or discovering novel categories. In this paper, we explore the relationship between logical reasoning and deep learning generalization in visual classification. A logical regularization termed L-Reg is derived which bridges a logical analysis framework to image classification. Our work reveals that L-Reg reduces the complexity of the model in terms of the feature distribution and classifier weights. Specifically, we unveil the interpretability brought by L-Reg, as it enables the model to extract the salient features, such as faces to persons, for classification. Theoretical analysis and experiments demonstrate that L-Reg enhances generalization across various scenarios, including multi-domain generalization and generalized category discovery. In complex real-world scenarios where images span unknown classes and unseen domains, L-Reg consistently improves generalization, highlighting its practical efficacy.