For this project, I used Singular Value Decomposition (SVD) to analyze two collections of facial images, and to prepare audio data for machine learning algorithms to identify music genres and artists.

The full paper describing this project is here.

For the first part, I analyzed two sets of faces from the Yale Face Database.  The first set included 2,414 images of human faces which were cropped to roughly the same size and position for each face.  The second set included 167 images of human faces which were uncropped.  I used SVD on each set of images.  I then compared a sample image with an increasing number of principal modes.

For the second part, I analyzed music from three genres: Jazz, Punk, and Baroque, including five artists of each genre.  I randomly selected samples of songs, took spectrograms of those samples, performed SVD on the spectrograms, and used the modes of the songs to train machine learning models and tested the models on other samples. I trained and tested a Naive Bayes model, and a multi-class ECOC model.