Hebbian Learning using Fixed Weight Evolved Dynamical ‘Neural’ Networks

In connectionist artificial neural network, the followers of the Hebbian learning (Hebb 1949) strongly believe that activities between two nodes can be increased or decreased by changing the connection weight between them. However, we claim and prove that learning can be produced without changing the connection weight in dynamic neural network as we believe that learning is produced by interaction with dynamic environment. To prove this I have reworked on research work of Izquierdo & Harvey (2007). However, I have achieved low fitness than their and my best evolved 4-node circuit is not as good as their in task achievement. I have used evolutionary approach to synthesize Continuous-Time Recurrent Neural Network parameters. The experimental methodology and output of the experiment have been described in details.

Download full pdf text