Towards automating active learning in collaborative bioprocess development

ERNESTO MARTÍNEZ

Berlin

Simposio; 7th BioProScale; 2022

Tecnical University of Berlin

The challenges of automating complex cognitive tasks in the framework of Industry 4.0/5.0 give rise to a key question: what is needed for next industrial revolution? More specifically, what means for collaborative bioprocess development the promise of autonomously gathering highly informative data sets using high-throughput robotic facilities and networked experimentation using ubiquitous digitization. In the talk it is argue that the next revolution is about automating learning, planning and model building alongside with creating cognitive interfaces for more skilled humans to interact with semi-autonomous robotic systems. The main enablers for automating active learning of all data needed throughout the development lifecycle are then highlighted. The pillars for next revolution are put forward as cognitive digital threads, using FAIR principles for data, workflows, experimental protocols and computational pipelines, and more importantly, active inference. Active inference is a powerful new theory in neuroscience that characterizes brain function prediction capability using mathematical formalisms and first principles that provide an entirely new approach to automate active learning in robotic platforms. Active inference is most appealing for high-throughput bioprocess development because it unifies state-estimation, control and bioreactor model learning as inference processes that are solved by optimizing a single objective functional: the free energy as it is used in variational Bayesian inference. Some preliminary results obtained for two case studies are presented.