The goal of this semester project is to provide baselines for genre classification, and maybe other problems in Music Information Retrieval (MIR) if time allows, on the FMA dataset (https://github.com/mdeff/fma). We will first test simple ML algorithms then devise and test Deep Networks. The idea is to show that the dataset is large enough for end-to-end learning, i.e. from the raw audio, with current DL techniques. This task involves the use of Python, the Jupyter notebook and libraries such as scikit-learn and Keras (TensorFlow backend).