Projects for the ECE class "Neural Networks and Intelligent Systems", 2020-2021
Students had to use 2 datasets from UCI (Machine Learning Repository) to apply classification techniques. We exprimented with the following classifiers: k-nearest neighbors, gaussian naive bayes, support vector machines, multilayer perceptron. We also hypertuned some of the models using grid search cross validation.
Students had to create 2 movie recommendation systems: the first using bag-of-words/cosine similarity metric and the second using Self Organising Maps (SOM's).
Image classification on Cifar-100 dataset using Deep Convolutional Neural Networks. We tested various optimazation techniques, model's architectures and used some state-of-art models using Transfer Learning.
Students had to create a smart agent capable of playing an Atari-2600 game using . Our game was named "Enduro". We used the following algorithms: PPO, A2C, DQN, QRDQN. Then we choose the one that achieved the best results and tries to optimize it further through hyperparameter tuning.