No abstract

No abstract

In the foreseeable future, synergistic advances in high-density memristive memory, scalable and massively parallel hardware, and neural network research will enable modelers to design large-scale, adaptive neural systems to support complex behaviors in virtual and robotic agents. A large variety of learning rules have been proposed in the literature to explain how neural activity shapes synaptic connections to support adaptive behavior. A generalized parametrizable form for many of these rules is proposed in a satellite paper in this volume [1]. Implementation of these rules in hardware raises a concern about the stability of memories created by these rules when the learning proceeds continuously and affects the performance in a network controlling freely-behaving agents. This paper can serve as a reference document as it summarizes in a concise way using a uniform notation the stability properties of the rules that are covered by the general form in [1].

No abstract

No abstract