Commonsense Reasoning and Knowledge Acquisition to Guide Deep Learning on Robots

Anonymous

Keywords: Knowledge Representation and Reasoning, Decision Tree Induction, State Constraints learning, Deep Learning

Abstract: Approaches based on deep network models are increasingly being used for pattern recognition and decision-making tasks in robotics and AI. These approaches are characterized by a large labeled dataset, high computational complexity, and difficultly in understanding the internal representations and reasoning mechanisms. As a step towards addressing these limitations, our architecture uses non-monotonic logical reasoning with incomplete commonsense domain knowledge, and inductive learning of previously unknown state constraints, to guide the construction of deep networks based on a small number of training examples. As an illustrative example, we consider a robot reasoning about the stability and partial occlusion of object configurations in simulated images of an indoor domain. Experimental results indicate that in comparison with an architecture based just on deep networks, our architecture improves reliability, and reduces the sample complexity and time complexity of training the deep networks.

Author Identity Visibility: Reveal author identities to reviewers