Solving Perception Uncertainty Problems in Robotics
Alberto Sanfeliu, Institut de Robòtica i Informàtica Industrial, Universitat Politècnica de Cataluyna, Spain
The Computational Magic of Pattern Recognition in Cortex: A Theory of Selectivity and Invariance
Tomaso Poggio, CBCL, McGovern Institute, Massachusetts Institute of Technology, United States
Alternating Direction Optimization for Imaging and Machine Learning Problems
Mário Figueiredo, Instituto de Telecomunicações and Instituto Superior Técnico, University of Lisbon, Portugal
Digital Pathology Imaging – The Next Frontier in Medical Imaging
Bikash Sabata, , United States
Solving Perception Uncertainty Problems in Robotics
Alberto Sanfeliu
Institut de Robòtica i Informàtica Industrial, Universitat Politècnica de Cataluyna
Spain
![]()
Brief Bio
Alberto Sanfeliu received the BSEE and PhD degrees from the Universitat Politècnica de Catalunya (UPC), Spain, in 1978 and 1982 respectively. He joined the faculty of UPC in 1981 and is full professor of Computational Sciences and Artificial Intelligence. He is director of the Institut de Robòtica i Informàtica Industrial, UPC-CSIC, director of the Artificial Vision and Intelligent System Group (VIS), former director of the UPC’s Automatic Control department and past president of AERFAI, (Spanish Association for Pattern Recognition).
He has worked on various theoretical aspects on pattern recognition, computer vision and robotics and on applications on vision defect detection, tracking, object recognition, robot vision, SLAM, robot navigation and urban robots. He has several patents on quality control based on computer vision. He has authored books in pattern recognition and SLAM, and published more than 230 papers in international journals and conferences. He has lead and participated in 34 R&D projects, 11 of them funded by the European Commission, and he has been the coordinator of the European project URUS (Ubiquitous Networking Robotics in Urban Areas). He is (or has been) member of editorial boards of several top scientific journals in computer vision and pattern recognition.
He received the prize to the Technology given by the Generalitat de Catalonia and is Fellow of the International Association for Pattern Recognition.
Abstract
Robotics has moved from industrial well controlled scenarios where the robots have an accurate knowledge of the objects, background and illumination, to scenarios where neither the environment conditions nor the objects are controlled. Perception systems are crucial to manage these issues; however they face with uncertainties which have not been considered in advance. Robots must perceive the environment and adapt to perception uncertainty when occurs. We will show some examples of these perception uncertainties problems in several real life robotics scenarios and the research approaches that have been adopted.
The Computational Magic of Pattern Recognition in Cortex: A Theory of Selectivity and Invariance
Tomaso Poggio
CBCL, McGovern Institute, Massachusetts Institute of Technology
United States
![]()
Brief Bio
Tomaso Poggio is Eugene McDermott Professor in the Department of Brain and Cognitive Sciences and at the Artificial Intelligence Laboratory.
He is also Co-Director of the Center for Biological and Computational Learning and was appointed Investigator immediately after the establishment of the McGovern Institute in 2000. He joined the MIT faculty in 1981, after ten years at the Max Planck Institute for Biology and Cybernetics in Tubingen, Germany. He received several awards such as the Otto-Hahn-Medaille Award of the Max-Planck-Society, the Max Planck Research Award (with M. Fahle), from the Alexander von Humboldt Foundation, the MIT 50K Entrepreneurship Competition Award, the 2003 Gabor Award, and the 2009 Okawa prize and 2009 Okawa prize.
Poggio is a Foreign Member of the Italian Academy of Sciences and a Fellow of the American Academy of Arts and Sciences.
Tomaso Poggio is one of the founders of computational neuroscience. He pioneered models of the fly’s visual system and of human stereovision,
introduced regularization theory to computational vision, made key contributions to the biophysics of computation and to learning theory,
developed an influential model of recognition in the visual cortex. He is one of the most cited computational neuroscientists.
Abstract
I conjecture that the sample complexity of object recognition is mostly due
to geometric image transformations and that a main goal of the ventral
stream is to learn-and-discount image transformations while preserving
sufficient selectivity. The theory predicts that the size of the receptive fields
determines which transformations are learned during development; that the
transformation represented in each area determines the tuning of the
neurons in the area; and that class-specific transformations are learned
and represented at the top of the ventral stream hierarchy. In problems of
pattern recognition, hierarchical, layered architectures -- similar to cortex --
may exploit in an optimal way unsupervised learning of transformations to
provide invariant and discriminative signatures to a supervised classifier.
Alternating Direction Optimization for Imaging and Machine Learning Problems
Mário Figueiredo
Instituto de Telecomunicações and Instituto Superior Técnico, University of Lisbon
Portugal
Brief Bio
Abstract
This talk will review recent work on the application of the alternating direction method of multipliers (ADMM) to several imaging inverse problems, as well as machine learning problems. It will be shown how ADMM provides an efficient and modular convex optimization tool, which allows addressing a variety of problems (namely, image restoration or reconstruction) using several different types of regularizers (total variation, frame-based analysis, frame-based synthesis), and formulations (constrained or unconstrained optimization). We will also review some recent uses of this class of methods in machine learning problems, namely for maximum a posteriori inference in graphical models.
In all the cases considered, the proposed methods inherit the theoretic convergence guarantees of ADMM and achieve state-of-the-art performance. To further illustrate the flexibility of this class of methods, we show how it can be used to seamlessly address other problems and formulations: hybrid analysis/synthesis regularization; group regularization (with or without group overlap); unmixing of hyperspectral data.
Digital Pathology Imaging – The Next Frontier in Medical Imaging
Bikash Sabata
United States
Brief Bio
Bikash Sabata is the Vice President for Imaging and Software, Digital Pathology at Roche Tissue Diagnostics (Ventana Medical Systems). His responsibilities encompass the development of novel pattern recognition and computer aided diagnosis algorithms in Digital Pathology to enterprise software systems that enable advanced workflow in the pathology lab and in the pathology practice. He has a strong combination of technology research in academic settings and product development in startups that span over 20 years of experience in developing imaging and computer vision architecture and algorithms. He most recently held the position of Chief Technology Officer at BioImagene that was acquired by Ventana/Roche. His key responsibilities included leading the software engineering team and the imaging team to build imaging algorithms, and the enterprise software product for Digital Pathology. Before BioImagene, he was the Co-founder and CTO at Aginova Inc, where his responsibilities included developing corporate vision, formation of the executive team, product architecture and development and revenue generation. In his roles as the cofounder of Primitive Root and the principal scientist at Peakstone Inc., he developed innovative applications of machine learning to distributed systems for predictive modeling and security. He was a scientist at SRI International and then at Information Extraction and Transport Inc where he led the R&D activities in Bayesian Network applications to Computer Vision, and Distributed Systems. Bikash was also an assistant professor at Wayne State University in Detroit and a researcher at Stanford University.
Bikash has many patents in the area of distributed information systems and digital pathology. His works on distributed and autonomous systems, learning in games, computer vision and parallel computing, have been widely published. He holds a Masters and a Ph.D in Electrical and Computer Engineering from the University of Texas at Austin. He has a Bachelor’s degree in Electrical Engineering from the Indian Institute of Technology (IIT) Bombay.
Abstract
Pathologists have practiced medicine relatively unchanged over the last century to render diagnosis of disease. However, over the last decade, the practice of Pathology is undergoing a foundational change.
In addition to the revolutions in diagnostic biomarkers, we are seeing a ground-swell in new imaging technologies. This new image based technology offers significant opportunities to the practice. However, it comes at a cost – there are technology, regulatory, and methodological challenges which may derail the adoption process if not addressed proactively and innovatively.
Pathology lags behind other medicine practice such as Radiology in the adoption of digital workflow. Recently, significant advances have been made in the capture and management of the whole slide images used in pathology practice and this is leading to an explosion in the data volume that completely eclipses the vast quantities of data being produced in Radiology. Another area of technology challenges is related to the analysis of the imagery data for detection, identification, recognition and quantification of the pathology in the slide. This clearly is the bleeding edge of digital pathology which will enable the creation of a paradigm shift in the practice of medicine.
This talk will focus on the technological challenges and briefly touch on the other areas. We will look at the different image capture technologies to the image analysis techniques. We will present the key computational challenges associated with this new image modality and computer-assisted image analysis to fully realize the potential of digital pathology in medical discovery and patient care.