Dear Colleagues,

In our paper we present Closed-Loop Deep Learning which combines the power of deep learning and classical control to create an efficient analogue multi-layered fast learning paradigm, for robotic navigation. At the heart of this algorithm is an error signal that arises from a reflex which is delicately used to both drive the closed loop system and train the deep learner simultaneously. Through mathematical derivation in z-space we show how to implement back-propagation in a closed-loop system.

Preprint:

https://arxiv.org/abs/2001.02970

Neural Computation (final version):

https://www.mitpressjournals.org/doi/abs/10.1162/neco_a_01317?journalCode=neco

Please do not hesitate to get in touch for a copy of the paper and to discuss your thoughts.

Warmest Regards,

Sama Daryanavard

Dr Bernd Porr

Biomedical Engineering Division, School of Engineering,

University of Glasgow, Glasgow G12 8QQ, U.K.