Noise-Tolerant Coreset-Based Class Incremental Continual Learning

2025/04/23 by Edison Mucllari, Mucllari, Edison, Aswin Raghavan +3
Computer Science · #Artificial Intelligence (cs.AI) #Computer Vision and Pattern Recognition (cs.CV) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Machine Learning (cs.LG) #Neural and Evolutionary Computing (cs.NE)

paper · pdf · doi:10.48550/arxiv.2504.16763

openalex publication_date 2025/04/23 · openalex created_date 2025/10/11 · openalex updated_date 2026/07/28

Abstract

Many applications of computer vision require the ability to adapt to novel data distributions after deployment. Adaptation requires algorithms capable of continual learning (CL). Continual learners must be plastic to adapt to novel tasks while minimizing forgetting of previous tasks.However, CL opens up avenues for noise to enter the training pipeline and disrupt the CL. This work focuses on label noise and instance noise in the context of class-incremental learning (CIL), where new classes are added to a classifier over time, and there is no access to external data from past classes. We aim to understand the sensitivity of CL methods that work by replaying items from a memory constructed using the idea of Coresets. We derive a new bound for the robustness of such a method to uncorrelated instance noise under a general additive noise threat model, revealing several insights. Putting the theory into practice, we create two continual learning algorithms to construct noise-tolerant replay buffers. We empirically compare the effectiveness of prior memory-based continual learners and the proposed algorithms under label and uncorrelated instance noise on five diverse datasets. We show that existing memory-based CL are not robust whereas the proposed methods exhibit significant improvements in maximizing classification accuracy and minimizing forgetting in the noisy CIL setting.

Citations

• Prioritized Generative Replay
• Persistent Backdoor Attacks in Continual Learning
• Maintaining Adversarial Robustness in Continuous Learning
• BrainWash: A Poisoning Attack to Forget in Continual Learning
• PACOL: Poisoning Attacks Against Continual Learners
• Adversary Aware Continual Learning
• Towards Adversarially Robust Continual Learning
• Robustness-preserving Lifelong Learning via Dataset Condensation
• Avalanche: A PyTorch Library for Deep Continual Learning
• Tackling Online One-Class Incremental Learning by Removing Negative Contrasts
• Lifelong Learning Metrics
• GCR: Gradient Coreset Based Replay Buffer Selection For Continual Learning
• Avalanche: an End-to-End Library for Continual Learning
• Adversarial Targeted Forgetting in Regularization and Generative Based Continual Learning Models
• Coresets via Bilevel Optimization for Continual Learning and Streaming
• Dark Experience for General Continual Learning: a Strong, Simple Baseline
• Targeted Forgetting and False Memory Formation in Continual Learners through Adversarial Backdoor Attacks
• Latent Replay for Real-Time Continual Learning
• Compacting, Picking and Growing for Unforgetting Continual Learning
• REMIND Your Neural Network to Prevent Catastrophic Forgetting
• Hidden Trigger Backdoor Attacks
• A continual learning survey: Defying forgetting in classification tasks
• EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks
• Three scenarios for continual learning
• Learn to Grow: A Continual Structure Learning Framework for Overcoming Catastrophic Forgetting
• Gradient Descent with Early Stopping is Provably Robust to Label Noise for Overparameterized Neural Networks
• Gradient based sample selection for online continual learning
• Experience Replay for Continual Learning
• Don't forget, there is more than forgetting: new metrics for Continual Learning
• Not All Samples Are Created Equal: Deep Learning with Importance Sampling
• Memory Aware Synapses: Learning what (not) to forget
• Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
• Lifelong Learning with Dynamically Expandable Networks
• Gradient Episodic Memory for Continual Learning
• Continual Learning with Deep Generative Replay
• Overcoming catastrophic forgetting in neural networks
• iCaRL: Incremental Classifier and Representation Learning
• Deep Residual Learning for Image Recognition
• Prioritized Experience Replay

Related

• Memory-Efficient Semi-Supervised Continual Learning: The World is its Own Replay Buffer
• Always Be Dreaming: A New Approach for Data-Free Class-Incremental Learning
• Recent Advances of Continual Learning in Computer Vision: An Overview
• A Multi-Head Model for Continual Learning via Out-of-Distribution Replay
• A Comprehensive Empirical Evaluation on Online Continual Learning