ACSE 8: Machine Learning¶
Module Lead: Dr. Lluis Guasch
Staff: Dr. Lluis Guasch
Course Description¶
The two main categories of Machine Learning methods, Supervised (Regression/ Classification) and Unsupervised Learning (Clustering / Dimensionality Reduction) are first introduced. Logistic Regression is presented in detail as it is a building block of Neural Networks. Logistic Regression also provides an opportunity to introduce the notations used in Deep learning.
For Supervised Learning, Feed-Forward Neural Networks and Back-Propagation are presented. For data distributed in space, whether 1-D (as a function of time), 2-D (images) or 3-D (material properties), Convolutional Neural Networks are more useful than Feed-Forward networks. They are introduced and examples of the most successful networks are presented and discussed.
For Unsupervised Learning, Principal Component Analysis (for Dimensionality Reduction) and k-Means (for Clustering) are first discussed. Then the rest of the course focusses on Generative Networks. Dimensionality Reduction with (Variational) Autoencoders or the generation of synthetic 1-D sequences, 2-D images or 3-D material properties with Generative Adversarial Networks (GANs) are introduced. This is also an opportunity to review the main probability concepts used in Deep Learning.
Students are asked to develop Deep Learning applications using Python and more specifically Pytorch.
Reading List¶
Pattern Recognition and Machine Learning by Christopher M. Bishop (Springer, 2006)
Deep Learning by Iann Goodfellow, Yoshua Bengio and Aaron Courville (The MIT Press, 2016)