Perceptron Consider the following pattern discrimination task: - First verify that this problem is not linearly separable. Try training a single-layer perceptron (SLP) network to see if it can solve the problem. The input vectors should be constructed from the above patterns by taking white boxes ( ) as 1 and black boxes ( ) as 1. - Can SLP network solve the given pattern discrimination task? Explain. - Then try a two-layer perceptron network. The input vectors should be constructed from the above patterns by taking white boxes () as 0.1 and black boxes ( size, learning rate parameter and target error used in the simulation? - Can MLP network solve the given pattern discrimination task? Explain. - How many hidden units are needed? What is the minimum number of hidden units necessary to solve this pattern discrimination task? - What patterns are recognized by the network if the following corrupted patterns are presented to the network? - How much noise (corruption) can your multi-layer perceptron (MLP) network handle? Support your answer by experimental/simulation results or figures when necessary. - Does the performance of the network improve if more units are used in the hidden layer? Explain