Keynote Lectures

Robust Face Recognition for Uncontrolled Settings
Harry Wechsler, George Mason University, United States, United States

Classifier Incongruence Detection for Anomaly Flagging in Machine Perception
Josef Kittler, University of Surrey, United Kingdom, United Kingdom

Deep-er Kernels
John Shawe-Taylor, Independent Researcher, United Kingdom, United Kingdom

Efficient and Versatile 3D Laser Mapping for Challenging Environments
Robert Zlot, Computational Informatics, , Australia, Australia

Robust Face Recognition for Uncontrolled Settings

Harry Wechsler
George Mason University, United States

Brief Bio
Harry Wechsler research focus covers adversarial learning, biometrics, change and anomaly / outlier detection, cyber security, identity management, interoperability, performance evaluation, re-identification, and multi-task learning. He organized and directed the seminal NATO Advanced Study Institute (ASI) on “Face Recognition: From Theory to Applications” (Sterling, UK, 1997), directed the design and development of FERET, and authored Reliable Face Recognition Methods (Springer, 2007). He is the author of 3 books, published over 300 scientific papers, and has 7 patents (together with his doctoral students). He is an IEEE Fellow and an IAPR (Int. Assoc. for Pattern Recognition) Fellow.

Abstract
The talk addresses the interplay between biometric challenges and the solutions proposed to develop and deploy robust face recognition applications that can operate under uncontrolled settings, adversarial encounters, and data (video) streaming conditions. Challenges include uncontrolled A-PIE (Age, Pose, Illumination, and Expression) settings, incomplete / masquerading information characteristic of denial and deception, outlier / imposter detection, varying data distributions, data fusion, and interoperability. The talk leverages image analysis / computer vision, machine and statistical {multi-task, ensemble, active} learning for face (object) recognition, re-identification, and surveillance / tracking applications. The solutions discussed are driven by regularization, compressive sensing and sparsity, similarity and rankings, and collective {multi-label, component-based, random sampling} classification methods. The talk concludes with an all encompassing view on biometrics that covers appearance, behavior, and cognitive state for the intertwined purpose of authentication, data mining, and situation awareness.

Classifier Incongruence Detection for Anomaly Flagging in Machine Perception

Josef Kittler
University of Surrey, United Kingdom

Brief Bio
Professor Josef Kittler FREng heads the Department of Electronics in the Faculty of Engineering and Physical Sciences. He received his BA, PhD and DSc degrees from the University of Cambridge. He joined the University of Surrey in 1986, after spending 5 years with EPSRC Rutherford Appleton Laboratory, and before that, conducting research on personal fellowships at Cambridge, Oxford, Southampton, and ENST Paris. He founded the Centre for Vision, Speech and Signal Processing, which now has over 100 researchers. He teaches and conducts research in Signal Processing and Machine Intelligence, with a focus on Biometrics, and Cognitive Vision. He published a Prentice Hall textbook on Pattern Recognition: A Statistical Approach, and more than 700 scientific papers. He is Series Editor of Springer Lecture Notes on Computer Science. He served as President of the International Association for Pattern Recognition 1994-1996. In 2006 he was awarded the KS Fu Prize from the International Association, for outstanding contributions to pattern recognition. He received Honorary Doctorates from the University of Lappenranta in 1999 and the Czech Technical University in Prague in 2007. In 2008 he was awarded the IET Faraday Medal and in 2009 he became EURASIP Fellow. He is a co-founder of OmniPerception Ltd.

Abstract

Machine perception systems are invariably complex. They involve multiple sensing modalities, many levels of representation, and hierarchical world models. The process of sensor(y) data interpretation typically engages a multitude of decision making mechanisms which successively convert sensor signals to symbols with a view of deriving full understanding of the input data. An essential function of machine perception is anomaly detection. Recently there has been considerable interest in developing sophisticated mechanisms of anomaly detection, aiming to emulate more closely the human perception system in its ability to distinguish nuances of anomaly. An important role in this context is classifier incongruence detection, as inconsistent decisions output by various decision-making processes (different modalities, contextual/noncontextual classifiers, generic/specific classifiers) are often indicative of something unusual. A key instrument in incongruence detection is a test statistics, which faithfully represents the concept of classifier incongruence. We argue that all existing measures of classifier incongruence are decision agnostic and consequently they suffer from responding to classifier incongruence with low signal to noise ratio. They are also are sensitive to estimation errors. We discuss a number of alternatives and demonstrate their favourable properties in experimental studies.

Deep-er Kernels

John Shawe-Taylor
Independent Researcher, United Kingdom

Brief Bio
John S Shawe-Taylor is a professor at University College London (UK) where he is Director of the Centre for Computational Statistics and Machine Learning (CSML). His main research area is Statistical Learning Theory, but his contributions range from Neural Networks, to Machine Learning, to Graph Theory. John Shawe-Taylor obtained a PhD in Mathematics at Royal Holloway, University of London in 1986. He subsequently completed an MSc in the Foundations of Advanced Information Technology at Imperial College. He was promoted to Professor of Computing Science in 1996. He has published over 150 research papers. He moved to the University of Southampton in 2003 to lead the ISIS research group. He has been appointed the Director of the Centre for Computational Statistics and Machine Learning at University College, London from July 2006. He has coordinated a number of European wide projects investigating the theory and practice of Machine Learning, including the NeuroCOLT projects. He is currently the scientific coordinator of a Framework VI Network of Excellence in Pattern Analysis, Statistical Modelling and Computational Learning (PASCAL) involving 57 partners.

Abstract
Kernels can be viewed as shallow in that learning is only applied in a single (output) layer. Recent successes with deep learning highlight the need to consider learning richer function classes. The talk will review and discuss methods that have been developed to enable richer kernel classes to be learned. While some of these methods rely on greedy procedures many are supported by statistical learning analyses and/or convergence bounds. The talk will highlight the trade-offs involved and the potential for further research on this topic.

Efficient and Versatile 3D Laser Mapping for Challenging Environments

Robert Zlot
Computational Informatics, , Australia

Brief Bio
Robert Zlot is a Senior Research Scientist at the CSIRO Computational Informatics division in Brisbane, Australia where he leads the Robotic Perception Team. His main research interests are in Field Robotics and 3D mapping. Robert obtained a PhD in Robotics from Carnegie Mellon University in 2006 where his research focused on coordinating teams of robots in mapping and exploration tasks. He joined CSIRO's Autonomous Systems Lab in 2007 where he has been developing algorithms and systems for mapping unstructured environments using laser range sensing. Since 2008, Robert has worked on continuous-time 3D trajectory estimation and non-rigid registration methods for continuously moving laser scanners. The solutions have been applied to a wide range of platforms, including mine-mapping and flying vehicles, as well as human-carried systems. Dr. Zlot is a co-inventor of the Zebedee 3D mapping system which is a handheld device capable of mapping an environment it is carried through. The system is currently being distributed commercially and has been utilized in a number of diverse applications worldwide. The technology has recently won several awards in Australia including the Eureka Prize for Innovative Use of Technology and the iAward for Research and Development.

Abstract
Continuous and mobile data acquisition is clearly more efficient than discrete and static data acquisition for mapping an environment, but poses challenges in terms of accurate positioning and data registration. Most commercial mobile mapping systems are reliant on Global Navigation Satellite Systems and high-grade inertial systems for positioning, and as a result have been primarily limited to street and aerial mapping applications. For more general applications, including indoors and underground, Simultaneous Localization and Mapping (SLAM) solutions present a methodology for modeling an environment without the need for installing additional positioning infrastructure. We have developed a continuous-time SLAM approach that can be applied to a wide variety of range sensor configurations and problem specifications, including laser scanners that sweep through their field of view relatively slowly compared to the platform's motion. The approach has been successfully demonstrated in mapping underground mines, forests, natural caves, building interiors, and cultural heritage sites around the world. A particular focus of this talk will be on the Zebedee handheld 3D mapping system, in which a laser scanner is mounted on a flexible spring, including some technical background on the underlying algorithms, and our experience in transferring the technology to a commercial market.