<ul><li>Machine learning algorithms often face the challenge of adapting models to concept drift, where task data distributions shift over time.</li><li>RCCDA is a dynamic model update policy that optimizes training dynamics while adhering to predefined resource constraints by utilizing past loss information and a tunable drift threshold.</li><li>Existing solutions for concept drift often have high computational overhead for resource-constrained environments and lack guarantees on resource usage or theoretical performance assurances.</li><li>Experimental results show that RCCDA outperforms baseline methods in inference accuracy while maintaining compliance with resource constraints under various concept drift scenarios, making it suitable for real-time machine learning deployments.</li></ul>

RCCDA: Adaptive Model Updates in the Presence of Concept Drift under a Constrained Resource Budget

Discover more