MATH 494: Mathematical Foundations of Machine Learning

Assignment #10

(Due Friday, May 17, 2024)

In this assignment, you will use a working script that performs simple classification of images using a neural network. The script generates a dataset of noisy 5x5 “images” of three types:

• happy smiley,
• sad smiley,
• neutral.

The Scikit-learn’s MLPClassifier is used to divide the dataset into training and testing partitions and to train the network to classify the “images.” The results of the testing are displayed (the training set score, the testing set score, and the confusion matrix). The noisiness is (clumsily!) implemented with thresholds of a uniform PRNG. The script, that will be emailed to you, is extremely simple and primitive

Your task is to rewrite fragments of the script to classify “images” into “noughts” and “crosses.” You are also supposed to experiment with the values of the thresholds and with the structure of the network (number and size of layers). The task will be explained in depth in class.

By May 17, you are supposed to email me the script and a short (half a page will be fine) write-up describing the results of your experiments with parameters.