In this project I have explored different topics in Deep Learning (such as network architecture, sparse models, sequential modelling, model combination, sequential training of different parts of the model and a bayesian approach) to predict the genre of a song with The Million Song Dataset.
The project (except the bayesian approach) has been developed as coursework for the Machine Learning Practical course at the University of Edinburgh (http://www.inf.ed.ac.uk/teaching/courses/mlp/index-2016.html). The data and a data provider to access the Million Song Dataset data have been provided by the teaching staff (more details in report.pdf).
The repository contains, in the Coursework folder, a report of the experiments undertaken in the coursework (report.pdf) as well as the implementation of these experiments (up to sequential modelling). The implementation of the experiments for Model Combination and the Correcting Model could be uploaded on request. Furthermore, in the Bayesian_Deep_Learning folder, I have implemented a Bayesian approach for the weights of the last layer of the model (Bayesian_Deep_Learning/EXPLANATION.md).