Cross-Sample Relational Fusion: Unifying Domain Generalization and Class-Incremental Learning

2026-05-09

AI Summary

This work addresses the dual challenges of catastrophic forgetting and cross-domain distribution shift commonly encountered in class-incremental learning (CIL) under real-world scenarios. To tackle these issues, we propose CrOss-sample Relational Fusion (CORF), a unified framework that jointly models domain generalization and CIL for the first time. CORF enhances both historical knowledge retention and cross-domain generalization through spatial contribution map-guided sample refinement and cross-sample multi-granularity relational distillation. The framework is designed to seamlessly integrate into existing CIL methods and demonstrates consistent performance gains across multiple benchmarks, effectively mitigating catastrophic forgetting while improving model robustness.

Abstract

Class-Incremental Learning (CIL) requires a learning system to learn new classes while retaining previously learned knowledge. However, in real-world scenarios such as autonomous driving, a system trained on urban roads in sunny weather may later need to operate in rural or highway environments with different traffic patterns and weather conditions. This requires the model not only to overcome catastrophic forgetting, but also to effectively handle domain shifts. In this paper, we propose CrOss-sample Relational Fusion (CORF), a unified framework to address domain shift and catastrophic forgetting simultaneously. To enhance generalizability, we perform selective refinement of training samples by leveraging spatial contribution maps to highlight semantically informative regions. Furthermore, we incorporate predictive confidence to adaptively weigh samples, thereby facilitating the learning of domain-agnostic representations. To alleviate forgetting, we propose a cascaded distillation framework that captures cross-sample relational dependencies across multiple feature hierarchies, enabling multi-grained knowledge transfer from previous tasks. CORF can be seamlessly integrated into existing CIL algorithms to enhance their generalizability, achieving competitive performance across various benchmark datasets. Code is available at https://github.com/LAMDA-CL/TMM26-CORF .

Problem

• Class-Incremental Learning
• Domain Generalization
• Catastrophic Forgetting
• Domain Shift

Innovation

• Cross-Sample Relational Fusion
• Domain Generalization
• Class-Incremental Learning
• Cascaded Distillation
• Domain-Agnostic Representation