Deep Learning for Visual Cognition

Abstract

The reliance on machine learning in cognitive systems is further justified as it is not always possible to equip an intelligent agent with prior knowledge. In the modern times, the field of machine learning has been dominated by deep learning based on artificial neural networks. This chapter introduces deep learning for visual cognition. It starts with a brief introduction to deep neural networks (DNN) and some basic reusable configurations to realize various learning algorithms. The chapter discusses various modes of learning realized through DNN in some illustrative computer vision tasks. It presents a broad overview of the architecture of the networks and the principles involved. The chapter focuses on DNN‐based implementations of visual attention models, which is crucial for alleviating information overload in cognitive systems. It presents some research on synthesizing Bayesian reasoning with DNNs, aimed at improving the robustness of inferences. Finally, the chapter concludes with some salient observations on the topic.