IDSIA has a very broad range of research interests, spanning most of Artificial Intelligence as it is understood today: machine learning, including deep learning/neural networks, control and signal processing, natural language processing, robotics, computer vision, search and optimisation, and more fundamental questions in uncertainty, probability, statistics, causal inference. To give an example, we have a 4-year Data project funded by the National Science Foundation as part of Switzerland's National Research Programme 75 "Big Data". In this project we deal with Gaussian processes, which can be understood as statistical neural networks, which can then provide uncertainty estimates relating to their own predictions – unlike traditional neural nets. This is very important in applications where we are evaluating risks. For example, a self-driving car needs to know whether the car's sensors are reliably warning of a potential accident ahead rather than a a person safely crossing the street.

David Duvenaud was collaborating on a project involving medical data when he ran up against a major shortcoming in AI. An AI researcher at the University of Toronto, he wanted to build a deep-learning model that would predict a patient's health over time. But data from medical records is kind of messy: throughout your life, you might visit the doctor at different times for different reasons, generating a smattering of measurements at arbitrary intervals. A traditional neural network struggles to handle this. Its design requires it to learn from data with clear stages of observation.

David Duvenaud was working on a project involving medical data when he hit upon a major shortcoming in AI. An AI researcher at the University of Toronto, he wanted to build a deep-learning model that would predict a patient's health over time. But data from medical records is kind of messy: throughout your life, you might visit the doctor at different times for different reasons, generating a smattering of measurements at arbitrary intervals. A traditional neural network struggles to handle this. Its design requires it to learn from data with clear stages of observation.