The network
- B0 - bias
- I1 - input
- 2<4 - recurrent connection of neuron 2
- 2 - hidden neuron
- O3 - output
In this project, the recurrent connections are explicitly separated in a different neuron. In the network above, the yellow neuron was added for the connection: 2 -> 2. Before each neural computation, the memory neurons get the value of their targets neurons. In this case, the 4th neuron (yellow), receives the value of the 2nd neuron.
The hidden neuron (2) has a Gaussian activation function.
Results
This project was inspired from the OpenAI Warmups exercises.
The agents are trained as the link suggests, the only difference it's that it uses only 100 instead of 100000 datasets.