Deep Statistical Relational Learning

Today, Artificial Intelligence permeates many aspects of our life. We interact with smartphones using our voice thanks to automatic speech recognition systems. We read the news from every country in the world in our language thanks to automatic translation systems. At the forefront of these developments is deep learning. This machine learning paradigm can process huge amounts of data to recognise useful patterns that will help future predictions.
However, deep learning is not suitable for tasks that cannot be reduced to pattern recognition and that, instead, require reasoning about objects and their relationships.  For example, writing a correct program given some specification is a task that will be hardly tackled by a deep learning model, no matter how much data is available, since programming requires reasoning and abstraction. The lack of such skills prevents deep learning to be considered a general recipe for Artificial Intelligence.
This research project aims at combining deep learning with reasoning techniques based on mathematical logic and probability theory. Deep learning will not be treated as a standalone paradigm but as an important ingredient of a larger theory. This challenge is a fundamental step towards systems that exhibit both pattern recognition and reasoning skills, leading to more general Artificial Intelligence.