Colloquium Cognitive Systems

Evaluation of Deep Learning in abstract classificatio

Assoz.-Prof. Antonio Rodríguez-Sánchez, PhD (Universität Innsbruck)

Abstract. Convolutional Neural Networks have become state of the art methods for image classification over the last couple of years. By now they perform better than human subjects on many of the image classification datasets. Most of these datasets are based on the notion of concrete classes (i.e. images are classified by the type of object in the image), that is, pattern recognition. In this talk I will introduce different image classification tasks, using abstract classes, which are to solve for humans, but variations of it are challenging for CNNs. The classification performance of popular CNN architectures is evaluated on these tasks and variations of the datasets that might be interesting for further research are identified. In this talk I will also investigate a promising deep learning architecture, that of capsule networks. I will present the limitations of capsule networks and a new training algorithm that can overcome them.

Bio. Antonio José Rodríguez Sánchez is currently an Associate Professor in the Intelligent and Interactive Systems group of the department of Computer Science at the Universität Innsbruck (Austria) leaded by Prof. Justus Piater. He was born in Santiago de Compostela (A Coruña), a beautiful city in the north-west of Spain. He completed his Ph.D. at the Center for Vision Research (York University, Toronto, Canada) on modeling attention and intermediate areas of the visual cortex under the supervision of John K. Tsotsos in 2010. He obtained the degree of M.Sc. in Computer Science at the Universidade da Coruña (Spain) in 1998. He received his B.Sc. in Computer Science at Universidad de Córdoba (Spain) with Honors. He did his Bachelor Thesis at the Université de La Rochelle (France). He has also finished 3 years of B.Sc. in Biology in the Universidad Autónoma de Madrid (Spain). His current research interests include computational neuroscience, (deep) neural networks, computer vision, machine learning and robotics.