PSO for NN Training

Here’s an example of a PSO optimizing the weights of a NN performing XOR.

Each particle is a 9D vector:

$$x=[w_1,w_2,w_3,w_4,b_1,b_2,b_3,v_1,v_2]$$

Our objective is to minimize the error between the desired and actual outputs of the network. This is done by randomly initializing a group of particles and letting them move in the search space until an acceptable error is reached.