Projects for the Statistical Modeling and Pattern Recognition course at the Technical University of Crete (Spring 2024).
Project 1 - Dimensionality Reduction and Bayesian Classification. PCA (including eigenfaces on a face dataset), a from-scratch LDA and a comparison with PCA, and multi-class LDA on the Iris dataset. Bayes decision theory, a Bayes classifier on hand-picked MNIST features, and minimum-risk classification.
Project 2 - Classifiers, Parameter Estimation and Clustering. The multi-class Perceptron algorithm, the logistic regression gradient, and Maximum Likelihood estimation of Gaussian parameters. Image compression with K-means and a Gaussian Mixture Model (GMM), plus a neural network built from scratch in NumPy / with TensorFlow-Keras.
Note:
mnist_train.csvis not included due to GitHub's 100 MB file limit. Download it from Kaggle and place it inProject 1/exercise1_5/data/.