FOCIL: Finetune-and-Freeze for Online Class Incremental Learning by Training Randomly Pruned Sparse Experts

2024-03-13

arXiv.org

AI Summary

Neural networks suffer from catastrophic forgetting in online class-incremental learning (Online CIL), and existing approaches rely on experience replay or model expansion—introducing high memory overhead and privacy risks. This paper proposes a replay-free dynamic sparse learning framework: for each incoming task, it generates a task-specific sparse subnetwork via random pruning, adaptively tunes sparsity ratio and learning rate, fine-tunes the subnetwork, and then freezes its weights to prevent interference. Crucially, no historical samples are stored, and the backbone architecture remains unmodified. To our knowledge, this is the first method achieving truly replay-free, low-overhead, near-zero-forgetting Online CIL. Extensive experiments demonstrate significant improvements over state-of-the-art methods on CIFAR-100 (10- and 20-task settings) and TinyImageNet (100-task setting). The code is publicly available.

Technology Category

• Machine Learning: Online Learning & Bandits
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Social networks and social learning
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Class incremental learning (CIL) in an online continual learning setting strives to acquire knowledge on a series of novel classes from a data stream, using each data point only once for training. This is more realistic compared to offline modes, where it is assumed that all data from novel class(es) is readily available. Current online CIL approaches store a subset of the previous data which creates heavy overhead costs in terms of both memory and computation, as well as privacy issues. In this paper, we propose a new online CIL approach called FOCIL. It fine-tunes the main architecture continually by training a randomly pruned sparse subnetwork for each task. Then, it freezes the trained connections to prevent forgetting. FOCIL also determines the sparsity level and learning rate per task adaptively and ensures (almost) zero forgetting across all tasks without storing any replay data. Experimental results on 10-Task CIFAR100, 20-Task CIFAR100, and 100-Task TinyImagenet, demonstrate that our method outperforms the SOTA by a large margin. The code is publicly available at https://github.com/muratonuryildirim/FOCIL.

Problem

• Prevent catastrophic forgetting in neural networks
• Enable online learning without data storage
• Handle real-world continual learning scenarios

Innovation

• Self-regulated neurogenesis for continuous learning
• Specialized concept cells prevent catastrophic forgetting
• Single over-parameterized network integrates multiple concepts