Short Intro: this is an advanced Master level course on machine and deep learning for human data analysis. Emphasis will be put on learning models and specifically on unsupervised learning techniques. The student will be trained on modern unsupervised data reduction and clustering algorithms, with deep learning models, including convolutional and recurrent neural networks. The teaching style will follow, at first, a presentation of the theoretical and technical tools, to then delve into their use within selected applications, their implementation and testing via extensive and carefully planned laboratory activity.
Selected topics in ML:
1. Unsupervised learning: Self Organising Maps, Neural Gas Networks, DBSCAN, denoising auto encoders (CNN and RNN-based).
2. Neural networks: convolutional neural networks (CNN), recurrent neural networks (RNN), batch normalisation, residual networks (ResNets), attention mechanisms (the transformer model).
Lab. activity: a key section of the course will be devoted to the lab. activity. Through this, the student will learn to implement the models and algorithms presented during the theoretical lessons. Special focus will be put on deep learning, including the implementation of convolutional (CNN) and recurrent neural networks (RNN). Various optimizations such as regularization techniques, and recent architectures such as autoencoders, residual networks (ResNets) and CNN inception layers will be discussed and implemented. The attention mechanism and the transformer model will be as well presented. Reference (tested) source code will be provided for each lab. lesson, which can be used by the student to develop their own projects both for this course and beyond.
Selected topics in ML:
1. Unsupervised learning: Self Organising Maps, Neural Gas Networks, DBSCAN, denoising auto encoders (CNN and RNN-based).
2. Neural networks: convolutional neural networks (CNN), recurrent neural networks (RNN), batch normalisation, residual networks (ResNets), attention mechanisms (the transformer model).
Lab. activity: a key section of the course will be devoted to the lab. activity. Through this, the student will learn to implement the models and algorithms presented during the theoretical lessons. Special focus will be put on deep learning, including the implementation of convolutional (CNN) and recurrent neural networks (RNN). Various optimizations such as regularization techniques, and recent architectures such as autoencoders, residual networks (ResNets) and CNN inception layers will be discussed and implemented. The attention mechanism and the transformer model will be as well presented. Reference (tested) source code will be provided for each lab. lesson, which can be used by the student to develop their own projects both for this course and beyond.
- Teacher: Michele Rossi