Storage Policy in Continual Learning: Different Approaches for Different Scenarios

Explore a comprehensive presentation on storage policy in continual learning, delivered by Julio Hurtado from the University of Pisa for the Data Learning working group. Delve into various approaches for different scenarios, beginning with an introduction to deep learning and its functionality. Examine training processes, distribution drifts, and methods for adapting to change. Compare human flexibility in learning to machine approaches in continual learning. Investigate replay-based approaches and storage policies, focusing on attributes of reliable memory systems. Learn about fast recovery techniques, high and low consistency, and the Consistency AWare Sampling (CAWS) method. Analyze limitations of CAWS and C-Score Proxies in continual learning. Study Class Incremental with Repetition (CIR) and its applications. Conclude with a summary of key points and participate in a Q&A session. Finally, gain insights into scenario generation for continual learning applications.