Invited Speakers

Computational Analysis in Cultural Heritage Applications

Tim Weyrich, University College London - UK

  • Thursday, August 28
  • 9:30 - 11:00
  • Room: FGV´s 12th floor Auditorium

Abstract: Through the increasing availability of high-quality consumer hardware for advanced imaging tasks, digital imaging and scanning are gradually pervading general practice in cultural heritage preservation and archaeology. In most cases, however, imaging and scanning are predominantly means of documentation and archival, and digital processing ends with the creation of a digital image or 3D model. At the example of two projects, the speaker will demonstrate how careful analysis of the underlying cultural-heritage questions allows for bespoke solutions that--through joint development of imaging procedures, data analysis and visualisations--directly support conservators and humanities researchers in their work. Tim Weyrich will report on his experiences with fresco reconstruction at the Akrotiri Excavation, Santorini, and on the reconstruction of fire-damaged parchment with London Metropolitan Archives.

Short-Bio: Tim Weyrich is an Associate Professor in the Virtual Environments and Computer Graphics group in the Department of Computer Science, University College London, and co-founder and Associate Director of the UCL Centre for Digital Humanities. Prior to coming to UCL, Tim was a Postdoctoral Teaching Fellow of Princeton University, working in the Princeton Computer Graphics Group, a post he took after having received his PhD from ETH Zurich, Switzerland, in 2006. His research interests are appearance modelling and fabrication, point-based graphics, 3D reconstruction, cultural heritage analysis and digital humanities.

Detectors and Descriptors for Three Dimensional Reconstruction of Real Scenes

Mario Campos, UFMG - Brazil

  • Thursday, August 28
  • 17:00 - 18:30
  • Room: FGV´s 12th floor Auditorium

Abstract: Three dimensional reconstruction of objects and scenes has been one of the key challenges for Computer Vision, and one of the first to be addressed by the community. Through the years impressive accomplishments have been attained by novel algorithms and innovative techniques using multiple two-dimensional images. Numerous new devices have also emerged with increased degree of resolution and accuracy. At the heart of many of such techniques is the ability to detect key points and to generate unique signatures which might enable consistent matching among images. More recently, the advent of commercially available, low cost devices has broadened the way and stirred even further the challenges of scene reconstruction by providing both image and depth information, even in real time. In this talk we will present recent approaches that use both information to build detectors and descriptors that are robust to several environmental conditions and at the same time are computationally efficient. We will show applications of such descriptors in the reconstruction of real indoor and outdoor scenes.

Short Bio: Mario Fernando Montenegro Campos, Ph.D., is a Professor of Computer Vision and Robotics in the Department of Computer Science at the Federal University of Minas Gerais (UFMG), Belo Horizonte, Brazil. He holds B.S. degrees in Engineering, and M.S. in Computer Science, all from UFMG, and a Ph.D. in Computer and Information Science from the University of Pennsylvania, USA. His research interests include cooperative robotics, robot vision, sensor information processing. His main contributions are in haptics, multirobot cooperation, aerial robotics and robot vision. He is the founder and director of the Vision and Robotics Lab -- VeRLab, UFMG, Brazil. He has been a Distinguished Lecturer in the IEEE Robotics and Automation Society.

Searching for Images to Measure the World

Robert Pless, Washington University in St. Louis - USA

  • Friday, August 29
  • 17:00 - 18:30
  • Room: FGV´s 12th floor Auditorium

Abstract: The world is observed by an enormous collection of cameras --- webcams, satellites and cell-phones. For 7 years, my group has explored what it takes to organize internet imagery as a tool for phenology, environmental, atmospheric and social measurement. Our approaches to analyzing this data set are inspired by a combination of time-lapse video artists Jason Salavon and Hiroshi Sugimoto and work to characterize the statistical invariants in images of natural scenes. When possible, we share our work as web-based tools or free apps so that more people can use Internet imagery to measure changes in their world. I will share some of our work to use these tools to measure the effects of weather changes on annual plant growth patterns, and characterizing how people use public spaces. I will also share unexpected successes we find from sharing these tools, including their use in tracking turtle migration and their forensic use to discover the location of a lost grave.

Short-Bio: Robert Pless is a Professor of Computer Science and Engineering at Washington University in St. Louis. His research focus is data driven approach to understanding motion and change in video, with a current focus on long term time-lapse imagery. Dr. Pless has a Bachelors Degree in Computer Science from Cornell University in 1994 and a PhD from the University of Maryland, College Park in 2000. He received the NSF CAREER award in 2006, chaired the IEEE Workshop on Omnidirectional Vision and Camera Networks (OMNIVIS) in 2003, and gave a keynote address at the IEEE Workshop on Application of Computer Vision in 2013.

Deep Learning That Just Works

James Bergstra, University of Waterloo - Canadá

  • Saturday, August 30
  • 14:00 - 15:30
  • Room: FGV´s 12th floor Auditorium

Abstract: Deep Learning has captured the imagination of researchers in both academia and industry after breakthrough empirical results in vision, speech processing, and natural language processing. However, the success of a deep learning algorithm is highly sensitive to myriad hyperparameters governing the data pre-processing, architecture, initial conditions, and learning. Reproducing these results and transferring them to new data sets is hard, even for experts, and insufficient exploration of hyperparameter configurations can lead to premature conclusions from successful models. New tools and model selection techniques based on Bayesian optimization and evolutionary search promise a more reliable approach to the design of deep learning systems, and machine learning systems more generally. This talk will describe software tools and ongoing research in computer-assisted design of deep learning systems for big data applications. Theano provides Graphics Processing Unit (GPU) code generation for computer-generated model configurations, so that deep networks can be trained in hours instead of days. Hyperopt is a black-box optimization library tailored for hyperparameter optimization by e.g. evolutionary search and Bayesian optimization algorithms. Together they demonstrate that algorithmic hyperparameter optimization represents a viable and fully-automated strategy for configuring deep learning architectures, and provide a new perspective on the success of deep learning.

Short-Bio: Dr. James Bergstra holds a Banting Postdoctoral Fellowship at the University of Waterloo Centre for Theoretical Neuroscience. His research interests include visual system models and learning algorithms, deep learning, Bayesian optimization, high performance computing, and music information retrieval. Previously he was a Research Scholar at the Rowland Institute for Science at Harvard University. He holds a doctoral degree from the University of Montreal with a dissertation on the use of complex cells models for deep learning. He co-authored Theano, a popular meta-programming system for Python that can target GPUs for high-performance computation.

I know what you are doing, and I can tell how well! — An outlook beyond (traditional) Activity Recognition.

WVHAR invited speaker: Thomas Ploetz, Newcastle University, UK

  • Saturday, August 30
  • 9:30 - 10:30
  • Room: FGV´s 12th floor Auditorium

Abstract: Computational Behaviour Analysis (CBA) is an emerging discipline that focuses on developing methods to study and understand behavioural phenotypes of humans with specific focus on health related applications. Examples of which are behaviour based screenings in Autism, skill assessment for Dementia patients or people with Parkinson's disease, or personalised rehabilitation support for stroke patients, to name but a few -- all with the objective of tailoring and improving individualised support. CBA is opportunistic regarding sensing modalities used for capturing human behaviour and apart from video based assessments Dr Ploetz and his team work with a multitude of sensors from the Pervasive and Ubiquitous computing domain. The focus of Ploetz's work is on developing and applying computational modelling techniques for CBA.

Short-Bio: Dr. Thomas Ploetz earned his PhD in Computer Science from Bielefeld University in Germany. His background is on machine learning and statistical pattern recognition. Currently he is an Assoc. Prof. (Senior Lecturer) at the School of Computing Science at Newcastle University in Newcastle upon Tyne, UK where the Machine Learning and Activity Recognition research activities within the Digital Interaction group at Culture Lab. Prior to his appointment in Newcastle he worked as a visiting Research Fellow at the Georgia Institute of Technology in Atlanta, USA, and as a post-doc in both Newcastle and TU Dortmund Universities (Dortmund, Germany).

Building a useful software dependency visualization system

WVIS invited speaker: Tim Dwyer, Monash University, Australia.

  • Saturday, August 30
  • 8:30 - 10:00
  • Room: FGV´s Cultural Centre

Abstract: I have been an information visualization and graph drawing researcher for the better part of 13 years so was therefore excited at the opportunity to build a "practical" graph-visualization tool with the weight and resources of a powerful software company behind me. However, I soon discovered that algorithms for layout and analysis are only one small piece in the puzzle to build a practical visualization tool that people will actually like and use. In this talk I hope to share that journey, some of the insights we acquired along the way, and---of course---spruik the tool itself: CodeMaps in the Microsoft Visual Studio 2013 IDE.

Short-Bio: Since 2012 Tim Dwyer is a Senior Lecturer and Larkins Fellow with the Faculty of IT at Monash University. From 2009 to 2012 he was a Senior Software Development Engineer with the Visual Studio group at Microsoft, USA. From 2008 to 2009 he was a Visiting Researcher with the VIBE group at Microsoft Research, USA. He was a Research Fellow at Monash University from 2005 to 2008 and received his PhD from the University of Sydney in 2005.

Interactive Visual Exploration and Analysis of Multi-Faceted Scientific Data

WVIS invited speaker: Helwig Hauser, University of Bergen, Norway.

  • Saturday, August 30
  • 11:00 - 12:30
  • Room: FGV´s Cultural Centre

Abstract: Modern scientific data, either from computational simulation, advanced measurements, or from complex modeling approaches, are increasingly often multi-faceted – we are thinking about multi-variate data, multi-dimensional data, multi-modal data, etc. The methodology of interactive visual analysis (IVA), which integrates computational methods with interactive processes, enables an iterative dialogue between the user and the data that makes it possible that both the unpaired perceptual and cognitive capabilities of the user are utilized as well as the steadily improving computational capabilities of modern computers. In this talk, we take a closer look at this interactive and iterative exploration and analysis loop and examine how a deeper understanding of data becomes possible through it. We do so out of an abstract perspective, while at the same time also considering a number of selected examples (from different application fields) which help to put the general considerations into a practical context.

Short-Bio: Helwig Hauser graduated in 1995 from Vienna University of Technology (TU Wien), Austria. In 1998, he finished his PhD project on the visualization of complex dynamical systems. In 2003, he finished his Habilitation at TU Wien, entitled ''Generalizing Focus+Context Visualization'' –in 2006 this work was awarded with the Heinz-Zemanek Award (given every 2nd year for exceptional works in computer science). In 2013, H. Hauser received the Dirk Bartz Prize for Visual Computing in Medicine from Eurographics. One of his main activities, more recently, is to chair visualization conferences, including EuroVis 2011, PacificVis 2012, and IEEE InfoVis 2013 and 2014. H. Hauser is member of the EuroVis Steering Com¬mit¬tee, the TopoInVis Steering Committee, and has served / is serving on the Editorial Boards of Computers & Graphics, Computer Graphics Forum, and IEEE Transactions on Visualization and Computer Graphics. After first working for TU Wien as assistant (since 1994) and later as assistant professor, he changed to the new VRVis Research Center in 2000. There, he led the basic research group on interactive visualization (until 2003) before he became the scientific director of VRVis (until 2007). Since then, 2007, he is a full professor in visualization at the University of Bergen in Norway, where he built up a new research group on visualization since.