Learning a synaptic learning rule

This paper presents an original approach to neural modeling based on the idea of searching, with learning methods, for a synaptic learning rule which is biologically plausible, and yields networks that are able to learn to perform diicult tasks. The proposed method of automatically nding the learning rule relies on the idea of considering the synaptic modiication rule as a parametric function. This function has local inputs and is the same in many neurons. The parameters that deene this function can be estimated with known learning methods. For this optimization, we give particular attention to gradient descent and genetic algorithms. In both cases, estimation of this function consists of a joint global optimization of (a) the synaptic modiication function, and (b) the networks that are learning to perform some tasks. The proposed methodology can be used as a tool to explore the missing pieces of the puzzle of neural networks learning. Both network architecture, and the learning function can be designed within constraints derived from biological knowledge.