9870 St Vincent Place, Glasgow, DC 45 Fr 45.
info@sitename.com
01234567890
BachecheDEI e siti degli insegnamenti: Various Master theses/internships on Machine Learning for Oil & Gas (in collaboration with Pietro Fiorentini SpA)
PROPOSTE DI TESI e/o STAGE
Various Master theses/internships on Machine Learning for Oil & Gas (in collaboration with Pietro Fiorentini SpA)
Various Master theses/internships on Machine Learning for Oil & Gas (in collaboration with Pietro Fiorentini SpA)
In collaboration with Pietro Fiorentini SpA https://www.fiorentini.com/it/
- Transfer Learning for Industry 4.0
Domain Adaptation and more in general Transfer learning (TL) are machine learning methods where a model developed for a task is reused as the starting point for a model on a second task. In the R&D scenarios it is quite common to develop multiple prototypes that differ in few details. Unfortunately, the direct application of a model, trained on one prototype to another, is not always straightforward.
In this work we propose the student to sample data from similar but different (also Arduino-based) prototypes and develop a strategy to solve the Transfer Learning task. - Active Learning for Regression tasks
Active Learning (AL) is a special case of machine learning in which an algorithm can interactively query a user to label new data points with the desired outputs. It is an extremely promising approach in situations where data labels/experiments are expensive and the user wants to minimize their number, for example in R&D scenarios. AL has been mostly applied to classification tasks and AL regression approaches are in their early stages.
In this thesis we propose to solve a specific regression task with the use of AL, with real experiments performed by the student using dedicated hardware (also Arduino-based). - Tiny Machine Learning for Industrial applications
Machine Learning models are often quite heavy compared to the limited resources of many devices. The most traditional way to overcome this issue is by using cloud-based services. However, this approach has some issues concerning data privacy, latency and energy consumption. A new approach, named TinyML, consists in techniques that allow a model to be computed on the edge.
This thesis has the goal to develop and test some these techniques, in particular, the student will be asked to create its dataset, the model and to deploy it on board on a dedicated micro-controller (also Arduino-based).
For additional information contact: gianantonio.susto@unipd.it