Sign and Relevance Learning

Sama Daryanavard, Bernd Porr

Standard models of biologically realistic or biologically inspired reinforcement learning employ a global error signal, which implies the use of shallow networks. On the other hand…