Task leads: Vladimir Loncar (lead), Michael Kagan (deputy), Maurizio Pierini (deputy)

This task focuses on development of software infrastructure for training efficient neural networks that will provide a standardized interface unifying neural network compression and optimization techniques such as pruning, quantization and neural architecture search. This work contributes to existing international projects; it will enable the development and deployment of hardware-optimal AI-based real-time algorithms at CERN, as described in WP2 and WP3.