Teaches artificial intelligence superhuman relational reasoning.
A key challenge in developing artificial intelligence systems with the flexibility and efficiency of human cognition is giving them a similar ability — to reason about entities and their relations from unstructured data. Solving this would allow these systems to generalize to new combinations of entities, making infinite use of finite means.
Modern deep learning methods have made tremendous progress solving problems from unstructured data, but they tend to do so without explicitly considering the relations between objects.
In two new papers, we explore the ability for deep neural networks to perform complicated relational reasoning with unstructured data. In the first paper — A simple neural network module for relational reasoning — we describe a Relation Network (RN) and show that it can perform at superhuman levels on a challenging task. While in the second paper — Visual Interaction Networks — we describe a general purpose model that can predict the future state of a physical object based purely on visual observations.
