In machine learning, we use data to automatically find dependences in the world, with the goal of predicting future observations. Most machine learning methods rely on statistics, and we will both discuss the theory underlying statistical learning as well as kernel machines building upon the theory.
However, statistical dependencies arise due to a more fundamental connection between variables: causality. Can causal knowledge help prediction in machine learning tasks? We argue that this is indeed the case, due to the fact that causal models are more robust to changes that occur in real world datasets. We provide some theoretical results as well as algorithmic approaches, and we touch upon the implications of causal models for machine learning tasks such as domain adaptation, transfer learning, and semi-supervised learning. Some of the causal problems are conceptually harder, however, the causal point of view can provide additional insights that have substantial potential for data analysis.
Time permitting, we will also present an application of causal modeling to the removal of systematic errors for the purpose of exoplanet detection.
(Empirical Inference Department Director, Max Planck Institute for Intelligent Systems, Tübingen);
(Research Scientist, Max Plank Institute for Intelligent Systems, Tubingen);
(Researcher at Facebook AI Research, Paris).
Morning: 3 hours/day lectures
Afternoon: 2 hours/day supervised tutorials as well as individual and team work.
Room and board
Accommodation is included in the registration fees.
Students will be hosted at the Guest House of Villa del Grumello, and in other facilities in the surroundings.
The organizers of the Summer School will take care of the reservations
Working days’ lunches are included in the registration fees, while dinners are not.
Attendance and final certificate
Full attendance of the activities of the summer school is mandatory for the participants
An attendance certificate will be awarded by Università Bocconi, subject to a positive participation to the program.