Continual Learning with Synthetic Boundary Experience Blending

2025-07-31

AI Summary

A primary cause of catastrophic forgetting in continual learning is the sparsity of stored samples in experience replay, which degrades decision boundaries. To address this, we propose Experience Blending—a novel framework that introduces synthetically generated boundary-proximal data as an implicit regularizer. Specifically, high-quality synthetic samples are batch-generated in a low-dimensional feature space using a multivariate differential privacy noise mechanism, then jointly trained end-to-end with real replayed samples. This approach effectively mitigates boundary simplification and enhances model stability. Empirical evaluation on CIFAR-10, CIFAR-100, and Tiny ImageNet demonstrates absolute accuracy improvements of 10%, 6%, and 13%, respectively, significantly outperforming nine state-of-the-art continual learning methods. Our results validate the critical role of synthetic boundary-aware data in experience replay for mitigating forgetting.

Technology Category

• Machine Learning: Life-Long and Continual Learning
• Search and Optimization: Learning to Search
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Practical large-scale studies of user experience
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Continual learning (CL) aims to address catastrophic forgetting in models trained sequentially on multiple tasks. While experience replay has shown promise, its effectiveness is often limited by the sparse distribution of stored key samples, leading to overly simplified decision boundaries. We hypothesize that introducing synthetic data near the decision boundary (Synthetic Boundary Data, or SBD) during training serves as an implicit regularizer, improving boundary stability and mitigating forgetting. To validate this hypothesis, we propose a novel training framework, {f Experience Blending}, which integrates knowledge from both stored key samples and synthetic, boundary-adjacent data. Experience blending consists of two core components: (1) a multivariate Differential Privacy (DP) noise mechanism that injects batch-wise noise into low-dimensional feature representations, generating SBD; and (2) an end-to-end training strategy that jointly leverages both stored key samples and SBD. Extensive experiments on CIFAR-10, CIFAR-100, and Tiny ImageNet demonstrate that our method outperforms nine CL baselines, achieving accuracy improvements of 10%, 6%, and 13%, respectively.

Problem

• Mitigate catastrophic forgetting in continual learning models
• Improve decision boundary stability with synthetic data
• Enhance experience replay using boundary-adjacent samples

Innovation

• Synthetic Boundary Data for boundary stability
• Multivariate DP noise for feature representation
• Experience Blending with stored and synthetic data