Deep Learning for Robot Perception and Cognition

Chapter 7: Lightweight deep learning

Paraskevi Nousi; Maria Tzelepi; Nikolaos Passalis; Anastasios Tefas    Department of Informatics, Aristotle University of Thessaloniki, Thessaloniki, Greece

Abstract

Deep convolutional neural networks have been excelling continuously on various challenging visual analysis tasks. Deep models with parameter-heavy architectures have been successfully trained and deployed on a multitude of applications, and owe their success partly due to the continuous development of increasingly more powerful graphical processing units. However, the power consumption and sheer size of such models hinder their applicability in robotics applications. Thus recent research has steered toward the optimization of deep learning architectures ...