Deep Learning with Logical Requirements

Owing to their remarkable ability of finding hidden patterns in data, deep learning models have been extensively applied in many different domains. However, recent works have shown that neural networks often fail to comply with requirements expressing background knowledge about the problem at hand. This represents a major drawback for deep learning models, as compliance with requirements is a necessary condition for standard software deployment. In this talk we will show how to build neural models that are not only guaranteed to be compliant with the given requirements over the output space, but are also able to learn from the background knowledge expressed by the requirements themselves and thus get better performance.  We will start by discussing simple hierarchical requirements and conclude with requirements expressed in full propositional logic.