Creating a neural net which is closely inspired by human brain. It has models for emotions, inhibitory/excitory, Long term potentiation/Depression. It reinforces a path positively for positive reinforcement and negatively for negative reinforcement.

It has two blocks - emotion and response Emotion block analyzes emotions and provide feedback to response block. Response block takes in input and emotion data and generates reponse.

It is a sparsely connected network with some structural connections and some random connections. The connections run both in forward as well as backward direction. So, any given layer will have connections both to previous and next layer in both forward and backward direction. Thus, any given layer would be processing combination of weighted previous & future data and current data. Each neuron has low threshold for firing as well as max weight. The threshold of each neuron firing will be adjusted by the emotion system. So the neurons can each have a different firing threshold. All the synapses have a low threshold below which no data will pass to next layer. This is to ensure that the synapses don't have extremely low weights. The neurons would always try to go towards average weight. So if a weight exceeds certain threshold the neuron try to decrease it to a more preferrable level while ensuring similar output. To make sure the output doesn't change, the ratio of weights for these sysnpases would remain same. This was needed as the system has model for Long term Potentiation, where more a synapse is used more its weight will increase. Also the synapses will receive some deterioration at every fixed interval. So if the system only recieve zero inputs(though rare) it's performance will degrade. To ensure that all synapses don't become "zero", low noise would be generated by system in event of long complete absolute silence. This will help it hallucinate/think/sleep

Since connections are random, it will not be readily optimized. The connections and weights are being optimized using Genetic algorithm.