Saving for the future: Enhancing generalization via partial logic regularization

2025-08-21

AI Summary

Generalization to unknown classes remains challenging in visual classification: class discovery methods are biased toward known classes; incremental learning suffers from catastrophic forgetting; and existing logical regularizers (e.g., L-Reg) rely on fully specified logical formulas, limiting adaptability to unseen categories. To address this, we propose Partial Logical Regularization (PL-Reg), the first framework to incorporate formal partial logic theories into visual learning. PL-Reg enables models to reserve expressive capacity for undefined logical formulas, relaxing the requirement of complete logical specifications inherent in conventional logical regularization. Evaluated across three tasks—generalized class discovery, multi-domain joint discovery, and long-tailed incremental learning—PL-Reg achieves consistent performance gains. Extensive experiments on multiple benchmarks demonstrate its superior adaptability to unknown classes and robust generalization capability.

Technology Category

• Machine Learning: Statistical Relational/Logic Learning
• Computer Vision: Visual Reasoning & Symbolic Representations
• Knowledge Representation and Reasoning: Logic Programming

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs

Abstract

Generalization remains a significant challenge in visual classification tasks, particularly in handling unknown classes in real-world applications. Existing research focuses on the class discovery paradigm, which tends to favor known classes, and the incremental learning paradigm, which suffers from catastrophic forgetting. Recent approaches such as the L-Reg technique employ logic-based regularization to enhance generalization but are bound by the necessity of fully defined logical formulas, limiting flexibility for unknown classes. This paper introduces PL-Reg, a novel partial-logic regularization term that allows models to reserve space for undefined logic formulas, improving adaptability to unknown classes. Specifically, we formally demonstrate that tasks involving unknown classes can be effectively explained using partial logic. We also prove that methods based on partial logic lead to improved generalization. We validate PL-Reg through extensive experiments on Generalized Category Discovery, Multi-Domain Generalized Category Discovery, and long-tailed Class Incremental Learning tasks, demonstrating consistent performance improvements. Our results highlight the effectiveness of partial logic in tackling challenges related to unknown classes.

Problem

• Enhancing generalization for unknown classes in visual classification
• Overcoming limitations of fully defined logical formulas in regularization
• Improving model adaptability to undefined logic formulas

Innovation

• Partial logic regularization for unknown classes
• Reserves space for undefined logic formulas
• Improves generalization via partial logic methods