SEAL: Searching Expandable Architectures for Incremental Learning

2025-05-15

AI Summary

To address the challenge of balancing catastrophic forgetting and resource efficiency in continual learning, this paper proposes a capacity-driven dynamic neural architecture search (NAS) framework. Unlike conventional task-wise fixed expansion NAS approaches, our method jointly optimizes both architectural topology and expansion strategy: (i) it employs differentiable NAS for end-to-end joint search; (ii) it introduces a lightweight capacity estimation metric to dynamically determine *whether* and *how* to expand the architecture; and (iii) it incorporates cross-task knowledge distillation to enhance stability. Evaluated on multiple continual learning benchmarks, the proposed method reduces average forgetting by 32–47%, improves final accuracy by 4.2 percentage points, and decreases parameter count by 38–61%, thereby significantly enhancing long-term performance and parameter efficiency.

Technology Category

• Search and Optimization: Learning to Search
• Machine Learning: Life-Long and Continual Learning
• 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
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

Incremental learning is a machine learning paradigm where a model learns from a sequential stream of tasks. This setting poses a key challenge: balancing plasticity (learning new tasks) and stability (preserving past knowledge). Neural Architecture Search (NAS), a branch of AutoML, automates the design of the architecture of Deep Neural Networks and has shown success in static settings. However, existing NAS-based approaches to incremental learning often rely on expanding the model at every task, making them impractical in resource-constrained environments. In this work, we introduce SEAL, a NAS-based framework tailored for data-incremental learning, a scenario where disjoint data samples arrive sequentially and are not stored for future access. SEAL adapts the model structure dynamically by expanding it only when necessary, based on a capacity estimation metric. Stability is preserved through cross-distillation training after each expansion step. The NAS component jointly searches for both the architecture and the optimal expansion policy. Experiments across multiple benchmarks demonstrate that SEAL effectively reduces forgetting and enhances accuracy while maintaining a lower model size compared to prior methods. These results highlight the promise of combining NAS and selective expansion for efficient, adaptive learning in incremental scenarios.

Problem

• Balancing plasticity and stability in incremental learning
• Reducing model expansion in resource-constrained incremental learning
• Dynamic architecture adaptation for efficient incremental learning

Innovation

• Dynamic model expansion using capacity estimation
• Cross-distillation training for stability preservation
• Joint NAS search for architecture and expansion policy