DIRA: Dynamic Domain Incremental Regularised Adaptation

Abstract

Autonomous systems (AS) often use Deep Neural Network (DNN) classifiers to allow them to operate in complex, high-dimensional, non-linear, and dynamically changing environments. Due to the complexity of these environments, DNN classifiers may output misclassifications during operation when they face domains not identified during development. Removing a system from operation for retraining becomes impractical as the number of such AS increases. To increase AS reliability and overcome this limitation, DNN classifiers need to have the ability to adapt during operation when faced with different operational domains using a few samples (e.g. 2 to 100 samples). However, retraining DNNs on a few samples is known to cause catastrophic forgetting and poor generalisation. In this paper, we introduce Dynamic Incremental Regularised Adaptation (DIRA), an approach for dynamic operational domain adaption of DNNs using regularisation techniques. We show that DIRA improves on the problem of forgetting and achieves strong gains in performance when retraining using a few samples from the target domain. Our approach shows improvements on different image classification benchmarks aimed at evaluating robustness to distribution shifts (e.g.CIFAR-10C/100C, ImageNet-C), and produces state-of-the-art performance in comparison with other methods from the literature.

References in corpus (17)

• Overcoming catastrophic forgetting in neural networks
• Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
• New insights and perspectives on the natural gradient method
• Alleviating catastrophic forgetting using context-dependent gating and synaptic stabilization
• Unsupervised Domain Adaptation through Self-Supervision
• Improving robustness against common corruptions by covariate shift adaptation
• On Quadratic Penalties in Elastic Weight Consolidation
• Evaluating Prediction-Time Batch Normalization for Robustness under Covariate Shift
• Distance-based Confidence Score for Neural Network Classifiers
• Task Agnostic Continual Learning Using Online Variational Bayes
• An Agency-Directed Approach to Test Generation for Simulation-based Autonomous Vehicle Verification
• Distance-Based Learning from Errors for Confidence Calibration
• Model Zoo: A Growing "Brain" That Learns Continually
• A Trustworthiness Score to Evaluate DNN Predictions
• Complete Agent-driven Model-based System Testing for Autonomous Systems
• Assessing Trustworthiness of Autonomous Systems
• Evaluation Metrics for DNNs Compression