Machine learning models adapt online for edge cloud IoT systems
Online Continual Learning in Edge-Cloud Applications
Machine LearningComputer Vision and Pattern Recognition
Summary
Machine learning models used in Internet of Things devices and cloud systems need to keep learning from new data while avoiding forgetting what they learned before. The problem is that these models often cannot access the original training data again and struggle when the new data looks very different. The authors present a method combining smart detection of unusual data with continual learning techniques that update models without erasing past knowledge. Their approach works across different fields like cars, healthcare, and farming, helping models adapt reliably over time.
What this means in practice
- •For automotive software engineers: Improve vehicle ML systems to adapt continuously to new driving conditions without losing previous knowledge.
- •For healthcare iot developers: Enable wearable or remote health devices to update their models in real time with patient data changes while avoiding model degradation.
Authors
Charalampos Davalas, Dimitrios Michail, Christos Diou, Iraklis Varlamis, Konstantinos Tserpes
Abstract
Deployed machine learning models in real-world IoT and edge/cloud systems face a critical reliability bottleneck: the requirement for online adaptation to non-stationary data streams while operating without access to the full original training dataset. Performance is severely degraded when models encounter Out-of-Distribution (OOD) inputs or suffer from catastrophic forgetting. Addressing these challenges requires adaptation mechanisms capable of learning from evolving data streams while preserving previously acquired knowledge. This paper proposes an edge-cloud methodology for integrating Continual Learning (CL) with Out-of-Distribution (OOD) Detection and Concept Drift Detection in IoT applications. The methodology incorporates ODIN and ECDD as alternative mechanisms for identifying informative samples for annotation and triggering rehearsal-based model updates. Experimental results across automotive, healthcare, and agricultural datasets demonstrate the applicability of the proposed methodology and its potential to support model adaptation while mitigating catastrophic forgetting.