Biases, Discrimination, and Fairness in Biometrics and Beyond
Julian Fierrez, Universidad Autonoma de Madrid, Spain
3D Motion Analysis with Event-based Sensors
Cornelia Fermüller, University of Maryland, United States
The Principle of Least Cognitive Action in Vision
Marco Gori, University of Siena, Italy
Biases, Discrimination, and Fairness in Biometrics and Beyond
Julian Fierrez
Universidad Autonoma de Madrid
Spain
http://atvs.ii.uam.es
![]()
Brief Bio
Julian FIERREZ received the MSc and the PhD degrees in telecommunications engineering from Universidad Politecnica de Madrid, Spain, in 2001 and 2006, respectively. Since 2004 he has been at Universidad Autonoma de Madrid, where he is Associate Professor since 2010. From 2007 to 2009 he was a visiting researcher at Michigan State University in the USA under a Marie Curie fellowship. His research is on signal and image processing, AI fundamentals and applications, HCI, forensics, and biometrics for security and human behavior analysis. He is actively involved in large EU projects in these topics (e.g., BIOSECURE, TABULA RASA and BEAT in the past; now IDEA-FAST, PRIMA and TRESPASS-ETN). Since 2016 he has been Associate Editor for Elsevier's Information Fusion and IEEE Trans. on Information Forensics and Security, and since 2018 also for IEEE Trans. on Image Processing. He has been General Chair of the IAPR Iberoamerican Congress on Pattern Recognition (CIARP 2018) and the Iberian Conference on Pattern Recognition and Image Analysis (IbPRIA 2019). Since 2020 he is a member of the ELLIS Society. Prof. Fierrez has received best papers awards at AVBPA, ICB, IJCB, ICPR, ICPRS, and Pattern Recognition Letters. He is also recipient of a number of world-class research distinctions, including: EBF European Biometric Industry Award 2006, EURASIP Best PhD Award 2012, Medal in the Young Researcher Awards 2015 by the Spanish Royal Academy of Engineering, and the Miguel Catalan Award to the Best Researcher under 40 in the Community of Madrid in the general area of Science and Technology. In 2017 he was also awarded the IAPR Young Biometrics Investigator Award, given to a single researcher worldwide every two years under the age of 40, whose research work has had a major impact in biometrics. [http://biometrics.eps.uam.es/]
Abstract
In the last few years, we are witnessing a growing interest in the Artificial Intelligence research community in studying bias effects when machine learning methods are applied on large amounts of data. These bias effects can stem from the data itself or from the learning process, which nowadays is clearly dominated by deep learning methods that most of the time are quite opaque. When those learning processes are related to AI applications dealing with personal information, or whose application affects people’s lives, then biases can result in unfair AI-based automated decision-making processes, very harmful in terms of undesired discrimination among population groups. This keynote will discuss the current state of the topic with special emphasis in AI applications involving face biometrics. Recent methods and approaches to reduce undesired discrimination towards fair biometrics will be also discussed.
3D Motion Analysis with Event-based Sensors
Cornelia Fermüller
University of Maryland
United States
![]()
Brief Bio
Cornelia Fermüller is a research scientist at the Institute for Advanced Computer Studies (UMIACS) at the University of Maryland at College Park. She holds a Ph.D. from the Technical University of Vienna, Austria and an M.S. from the University of Technology, Graz, Austria, both in Applied Mathematics. She co-founded the Autonomy Cognition and Robotics (ARC) Lab and co-leads the Perception and Robotics Group at UMD. She is the PI of an NSF-sponsored Science of Learning Center Network for Neuromorphic Engineering. Her research is in the areas of Computer Vision, Human Vision, and Robotics. She studies and develops biologically inspired Computer Vision solutions for systems that interact with their environment. In recent years, her work has focused on the interpretation of human activities, and on motion processing for fast active robots (such as drones) using as input bio-inspired event-based sensors.
http://users.umiacs.umd.edu/users/fer
Abstract
Event-based sensors have gained increased popularity in the fields of Computer Vision and Robotics because they offer exciting alternatives for motion perception. These neuromorphic imaging devices, inspired by the transient pathway of mammalian vision, record at very high temporal resolution the changes in the scene. In this way, they produce a continuous stream of events with each pixel operating asynchronously and independently, which allows us to interpret continuous motion. The data’s unique properties (high dynamic range and temporal resolution, low latency and bandwidth) are valuable for solving the tasks of the early visual motion pathway - 3D motion estimation, segmentation and tracking robustly in even challenging scenarios.
Treating the data as point clouds - the so-called event clouds in x-y-t space, we developed new constraints, developed algorithms, and demonstrated them in robotics applications. The constraints relate the event clouds and their surface normal vectors to 3D motion, and the shape of the clouds to scene geometry and object motions. These constraints are used by aligning clouds using time and space information, and implemented in classical optimizations as well as deep neural networks. Finally, new datasets for evaluating 3D motion, structure, and depth were collected, and autonomous driving as well as drone applications for tracking, dodging and pursuit were developed.
The Principle of Least Cognitive Action in Vision
Marco Gori
University of Siena
Italy
Brief Bio
Marco Gori received the Ph.D. degree in 1990 from Università di Bologna, Italy, working partly at the School of Computer Science (McGill University, Montreal). In 1992, he became an Associate Profes-sor of Computer Science at Università di Firenze and, in November 1995, he joint the Università di Siena, where he is currently full profes-sor of computer science, where he is leading the Siena Artificial Intelli-gence Lab (SAILAB) http://sailab.diism.unisi.it/
Professor Gori is primarily interests in machine learning with ap-plications to pattern recognition, Web mining, game playing, and bioin-formatics. He has recently published the monograph “Machine Learning: A constraint-based approach,” (MK, 560 pp., 2018), which contains a unified view of his approach. His pioneering role in neural networks has been emerging especially from the recent interest in Graph Neural Net-works, that he contributed to introduce in the seminal paper “Graph Neural Networks,” IEEE-TNN, 2009, which received nearly 500 cita-tions in 2019.
Professor Gori has been the chair of the Italian Chapter of the IEEE Computation Intelligence Society and the President of the Italian Asso-ciation for Artificial Intelligence. He is a Fellow of IEEE, a Fellow of EurAI, and a Fellow of IAPR. He was one the first people involved in European project on Artificial Intelligence CLAIRE, and he is currently a Fellow of Machine Learning association ELLIS. He is in the scientific committee of ICAR-CNR and is the President of the Scientific Commit-tee of FBK-ICT. He has been recently invited by the Agence Nationale de la Recherche of France to be a member of the pool of experts for the French national research plan on AI. Dr. Gori is currently holding the 3IA Chair position at the Université Cote d’Azur.
Abstract
In this talk we introduce the principle of Least Cognitive Action with the purpose of understanding perceptual learning processes, with special emphasis on vision. The principle closely parallels related approaches in physics, and suggests to regard neural networks as systems whose weights are Lagrangian variables, namely functions depending on time. Interestingly, neural networks “conquer their own life” and there is no neat distinction between learning and test; their behavior is characterized by the stationarity of the cognitive action, an appropriate functional which contains a potential and a kinetic term. While the potential term is somewhat related to the loss function used in supervised and unsupervised learning, the kinetic term represents the energy connected with the velocity of weight change. Unlike traditional gradient descent, the stationarity of the cognitive action yields differential equations in the connection weights, and gives rise to a dissipative process which is needed to yield ordered configurations. We give conditions under which this learning process reduces to stochastic gradient descent and to Backpropagation. We give examples on supervised and unsupervised learning, and briefly discuss the application to deep convolutional neural networks, where an appropriate Lagrangian term is used to enforce motion invariance in the visual feature extraction.