I am part of the Machine Learning Group at the Department of Mathematics, Università degli studi di Padova.
If you're a Master student (studente della laurea magistrale) in Computer Science (or in Data Science) and you're looking for a thesis in Machine learning, you're in the right place!
In the Publications page, you can find some research papers, that you can read, that represent the state-of-the-art in their respective areas.
You can get inspired by those works, and discuss with us your ideas, until you'll get to an understanding deep enough to let you contribute in that specific topic and to write your thesis on it.
On the other hand, if you prefer a more "guided" path, you can contact me and I'll present you some ideas!
You'll be supervised by me and prof. Alessandro Sperduti.
In the following, some possible starting points for your thesis (I'll try to keep this list updated).
Comparison of different GCNs with an appropriate model selection.
Convolutional Neural Networks for video analysis.
Application of Reservoir Computing techniques (e.g. Echo State Networks) to a Business Process Mining problem of regression over sequences.
Implementation and comparison of several neural models for learning on graphs.
Code refactoring, documentation and publication of a Python library compatible with scikit-learn for learning on structured data.
Space-efficient representation of categorical features (i.e. non-numerical features, like words) for learning algorithms, e.g. Neural Nets or SVM.
Dimensionality reduction: comparing different dimensionality reduction techniques, to make the learning process more efficient.
Dimensionality expansion (feature generation): the idea is to augment the samples' dimensionality (e.g. ELM or "Weighted sums of random kitchen sinks") , in order to apply in the new space efficient learning algorithms for learning in the streaming setting / anytime learning.
Evolution of Graph Node Kernels (extension of the paper "Approximated Neighbors MinHash Graph Node Kernel").