The project consisted of implementing a Multilayer perceptron in Matlab using standard libraries to recognize handwritten digits.
The goal was to find the optimal number of hidden layers and nodes per layer to minimize validation error then use these parameters to report testing error on a unseen testing set
The Multilayer perception implementation included an input layer with node corresponding to the total pixels per each image, and output layer of 10 nodes corresponding to which value the perceptron believed the handwritten digit was, and a variable number of hidden layers and nodes believed to capture edges and characteristics unique to each handwritten number.