Continual learning via probabilistic exchangeable sequence modelling

2025-03-26

AI Summary

Continual learning (CL) faces challenges including catastrophic forgetting, inadequate uncertainty quantification, and high computational overhead. This paper proposes CL-BRUNO—the first probabilistic CL framework integrating exchangeable sequence modeling with neural processes, supporting both task- and class-incremental learning without storing historical data, thereby enabling scalable Bayesian updates. Its core innovation lies in introducing an exchangeability prior that jointly enforces distribution-level and function-space regularization, unifying the treatment of diverse CL scenarios. CL-BRUNO simultaneously ensures privacy preservation, minimal storage requirements (retaining only parameter-level summaries), and principled output uncertainty quantification. Extensive experiments on natural image and biomedical benchmarks demonstrate that CL-BRUNO significantly outperforms state-of-the-art methods, validating its effectiveness, generalization capability, and robustness in real-world decision-making settings.

Technology Category

• Machine Learning: Life-Long and Continual Learning
• Reasoning under Uncertainty: Probabilistic Programming
• Natural Language Processing: Learning & Optimization for NLP

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems

Abstract

Continual learning (CL) refers to the ability to continuously learn and accumulate new knowledge while retaining useful information from past experiences. Although numerous CL methods have been proposed in recent years, it is not straightforward to deploy them directly to real-world decision-making problems due to their computational cost and lack of uncertainty quantification. To address these issues, we propose CL-BRUNO, a probabilistic, Neural Process-based CL model that performs scalable and tractable Bayesian update and prediction. Our proposed approach uses deep-generative models to create a unified probabilistic framework capable of handling different types of CL problems such as task- and class-incremental learning, allowing users to integrate information across different CL scenarios using a single model. Our approach is able to prevent catastrophic forgetting through distributional and functional regularisation without the need of retaining any previously seen samples, making it appealing to applications where data privacy or storage capacity is of concern. Experiments show that CL-BRUNO outperforms existing methods on both natural image and biomedical data sets, confirming its effectiveness in real-world applications.

Problem

• Addresses computational cost and uncertainty in continual learning
• Provides scalable Bayesian updates for real-world decision-making
• Prevents catastrophic forgetting without retaining past data samples

Innovation

• Probabilistic Neural Process-based continual learning model
• Deep-generative unified probabilistic framework
• Distributional and functional regularization against forgetting