Day 1 |
Pinhas Ben-Tzvi, Ph.D.
Department of Mechanical and Aerospace Engineering,
The George Washington University, Washington, DC
Symbiosis of Mobile Robotic Locomotion and Manipulation
on Rough Terrain
The prevailing design paradigm of state-of-the-art mobile robots consists of a stand-alone manipulator arm attached on top of a locomotion platform. The locomotion platform provides mobility and the arm provides manipulation, each with distinct functionalities that are not interchangeable. This approach renders the overall system’s functionally very limited for rough terrain applications. This talk presents a novel hybrid mechanism mobile robot system whereby the locomotion and manipulation platforms are designed as one entity to support both locomotion and manipulation symbiotically. The design paradigm is described as well as a novel generalized control hardware architecture based on embedded on-board wireless communication network between the robot’s subsystems. This approach resulted in modular control architecture, wireless network between subsystem’s actuators and sensors, and operational fault-tolerance. The new design paradigm and control hardware architecture demonstrate the qualitative and quantitative performance improvements of the mobile robot in terms of the new locomotion and manipulation capabilities it provides. Experimental results are presented to demonstrate new operative tasks that the robot was able to accomplish, such as traversing challenging obstacles, and manipulating objects of various capacities; functions often required in various challenging applications, such as search & rescue missions, hazardous site inspections, and planetary explorations.
Biography
Pinhas Ben-Tzvi received the B.Sc. degree (Summa Cum Laude) from the Technion – Israel Institute of Technology, Haifa, Israel, in 2000, and the M.Sc. and Ph.D. degrees from the University of Toronto, Toronto, ON, Canada, in 2004 and 2008, respectively, both in mechanical engineering. From 2000 to 2002, he was employed by General Electric Medical Systems, where he worked as an R&D Engineer on the development and integration of medical diagnostic robotic and mechatronic systems (CT, PET, and PET/CT). Dr. Ben-Tzvi is currently an Assistant Professor in the department of Mechanical and Aerospace Engineering and the Founder and Director of the Robotics and Mechatronics Laboratory at the George Washington University. His areas of research and academic interests are focused in robotics, mechatronics, intelligent autonomous systems, mobile robotic locomotion and manipulation, and smart materials-based development of novel sensors and actuators for biomedical and miniature mechatronic and microrobotic systems. Dr. Ben-Tzvi is a member of the American Society of Mechanical Engineers (ASME) and the Institute of Electrical and Electronics Engineers (IEEE).
|
Simon X. Yang, PhD., Professor
Advanced Robotics & Intelligent Systems (ARIS) Laboratory
School of Engineering, University of Guelph
Intelligent monitoring and analysis of livestock farm odour
using wireless e-nose network
The livestock farm odor is very complex, which constitutes up to 200 odorous chemical components that interact nonlinearly with each other. Odor complaints are a significant issue for today's livestock industries. Odor control technologies are greatly impeded by the lack of science-based approaches to assess the levels and effects of odors. In this study, an electronic nose (e-nose) system consisting of a sensor array, a signal processing system, and an intelligent analysis software suite was developed. An innovative neuro-fuzzy inference system is developed to calibrate the e-nose reading to the olfactory perception of human panelists. The developed e-nose system is capable of providing a reliable and objective odor measurement with low operation cost and high consistency. In addition, to have a real-time overall profile of the odor around a livestock farm, a wireless e-nose network was developed to obtain the dynamic odor dispersion plume around the farm. A novel biologically inspired odor dispersion model is developed to dynamically display the odor dispersion plume around livestock farms, and to facilitate effective and efficient odour control practice.
Biography
Prof. Yang received the B.Sc. degree in Engineering Physics from Beijing University, China, in 1987, the first of his two M.Sc. degrees in Biophysics from Chinese Academy of Sciences, Beijing, China, in 1990, the second M.Sc. degree in Electrical Engineering from the University of Houston, USA, in 1996, and the Ph.D. degree in Electrical and Computer Engineering from the University of Alberta, Edmonton, Canada, in 1999. He joined the University of Guelph in Canada in August 1999. Currently he is a Professor and the Head of the Advanced Robotics & Intelligent Systems (ARIS) Laboratory at the University of Guelph. Prof. Yang’s research expertise includes Intelligent Systems, Robotics, Sensors and Signal Processing, Multi-sensor Fusion, Wireless Sensor Networks, Intelligent Control and Computational Neuroscience. Dr. Yang has served as an Associate Editor of IEEE Transactions on Neural Networks, IEEE Transactions on Systems, Man, and Cybernetics, Part B, International Journal of Robotics and Automation, and several other international journals. He has involved in the organization of many international conferences. He was the General Chair of 2011 IEEE International Conference of Automation and Logistics.
|
Ryan Gariepy
Co-Founder and CTO, Clearpath Robotics
Towards Ubiquitous Mobile Robots
The deployment of autonomous systems to commercial applications is increasing. Wide swaths of traditional industries are accepting mobile robots, first responders are seeking unmanned systems for their vehicles, and autonomous robots are entering the home. With the spread of this technology into the hands of those who have never seen a robot before comes an increased demand for systems which are safe, predictable, and easy to use. How do we design systems which are useful outside of the lab?
This talk lays out key challenges which stand in the way of wider adoption of the robotics technology we already possess, including system robustness, user interfaces, regulatory aspects, and deployment logistics. It explains how increasing the capability of mobile robots can in fact decrease their usability.
A design strategy will be presented which considers these challenges at the onset. Examples of failed, successful, and promising applications for autonomous robotics across a variety of problem domains and industries will be discussed, ranging from military ordinance disposal to automated civil infrastructure inspection.
Biography
Ryan Gariepy has been focused on the field of unmanned systems since he began his engineering studies. Prior experience at robotics companies such as Kiva Systems and Aeryon Labs has convinced him that the time is right for the widespread commercial use of autonomous systems. He believes that robotics can be made as easy to use and reliable as any other tool, and is driving this vision in his current role at Clearpath Robotics. Mr. Gariepy is currently working on further expanding Clearpath's research & educational tools as well as guiding the development of their autonomous surveying technology. He completed both a B.A.Sc. degree in Mechatronics Engineering and a M.A.Sc. degree in Mechanical Engineering at the University of Waterloo. He was recently an invited panelist at the RoboBusiness Leadership Summit, and is a primary organizer of the Robot Operating System developers' conference.
|
Day 2 |
Masoud Makrehchi
Research Scientist
Thomson Reuters
Social Link Prediction
Web 2 has revolutionized our daily online experience and had a significant impact on the information society. The most important aspect of the Web 2 is that it is social. People are not only consume the information generated and presented on the Web, they also actively participate in generating, transforming, and propagating information. What makes the Web 2 and social media very powerful is the underlying social structure and communities in every single online service.
On the other hand, having the Usenet experience, we know if communities and social networks fail in having a sustainable growth over time, they will eventually die. In addition to that, network growth is essential for better user experience by which we can make the communities stickier. In order to have stickier communities, online social networks has to constantly recommend appropriate social links to the members.
In this talk, several strategies for link recommendation and prediction are discussed. First, a taxonomy of the topic is presented. Then, social data and features are briefly discussed. Several link prediction methods including (semi)supervised and unsupervised techniques, and also link prediction using hidden topics are detailed. Finally, some applications and examples are outlined.
Biography
Masoud Makrehchi received his Ph.D. in Electrical and Computer Engineering from University of Waterloo, in 2007. He joined Thomson Reuters R&D department based in Greater Saint Paul-Minneapolis Area, MN, in May 2008 as research scientist. His current primary research interests are in the areas of recommender systems, text and data mining, machine learning, social computing, and mining social data.
|
Wayne Oldford
University of Waterloo
Visual Clustering of High-dimensional Data
by Navigating Low-dimensional Spaces
The structure of a set of high dimensional data objects (e.g. images, documents, molecules, genetic expressions, etc.) is notoriously difficult to visualize. In contrast, lower dimensional structure (esp. 3 or fewer dimensions) is natural to us and easy to visualize. A not unreasonable approach, then, is to explore one low dimensional visualization after another in the hope that, together, these will shed light on the higher dimensional structure. A familiar example is the parallel coordinate plot, where a sequence of one-dimensional projections are connected to provide insight into the structure of a high dimensional data set.
In this talk, I describe a graph theoretic structure that represents low-dimensional spaces as graph nodes and transitions between spaces as edges. Of interest, are walks along these graphs that reveal meaningful structure. If the nodes are one-dimensional, a walk corresponds to a parallel coordinate plot; if they are two dimensional and edges exist, say, only between 2d spaces which share a variate, then the walk could be represented dynamically as a series of scatterplots, one transitioning into the next via a 3d rigid transformation.
These graphs can be used to dynamically explore high dimensional data to visually reveal cluster structure. RnavGraph is a tool developed by Adrian Waddell and myself for that purpose. This visualization tool can be used for visual cluster analysis in place of, or in concert with, automated methods. I will demonstrate this by a quick analysis of a data set from food chemistry which contains eight measurements of fatty acid concentrations on 572 Italian olive oils.
These graphs, unfortunately, grow in size with the square of the number of dimensions. Fortunately, there are numerous means for constructing only the more interesting regions of each graph. Some restrictions are imposed by the statistical context, others by empirical measures on the data itself. Of the latter, scatterplot diagnostics (scagnostics) are especially valuable. There are also numerous dimension reduction methods (including manifold learning methods) which can be used in conjunction with the graph reduction methods. The “Frey image” data is used to demonstrate how images can be analysed using RnavGraph.
Biography
Wayne Oldford earned an Honours B. Math. degree (double major - Statistics; Combinatorics & Optimization) from the University of Waterloo in 1977, an M.Sc. and Ph.D. in Statistics from the University of Toronto in 1979 and 1982, respectively. He worked as a Survey Methodologist at Statistics Canada from 1977 to 1979. From 1982 to 1986, he was a principal researcher at the Center for Computational Research in Economics and Management Science in the Sloan School of Management at MIT. Wayne joined the Department of Statistics and Actuarial Science at the University of Waterloo in 1986 and has been cross-appointed to the David R. Cheriton School of Computer Science since 1998. He has held various administrative positions within the University including Director of the Centre for Computational Mathematics in Industry and Commerce, Associate Dean for Computing for the Faculty of Mathematics, Director of the Mathematics Faculty Computing Facility, and Associate Chair of the Department of Statistics and Actuarial Science. He has served as an Associate Editor of JASA, JCGS, Statistics and Computing, The International Statistical Review and as Editor-in-Chief of Statistics and Computing from 2001 to 2006. He has co-edited four volumes of "Statistics, Science, and Public Policy". He has served on several program committees including the workshops on Artificial Intelligence and Statistics, and Intelligent Data Analysis and was one of three founders of the Society for AI and Statistics which manages the biennial AISTATS workshops. Wayne has been an elected member of the Board of Directors of the Statistical Society of Canada and Member of Council for the International Association for Statistical Computing. He is an elected member of the International Statistical Institute, member of several statistical societies, and serves as a research advisor to Primal Fusion, a local internet knowledge based company.
|
Ruslan Salakhutdinov
University of Toronto
Learning Hierarchical Models
Building intelligent systems that are capable of extracting meaningful
representations from high-dimensional data lies at the core of solving
many Artificial Intelligence tasks, including visual object recognition,
information retrieval, speech perception, and language understanding.
Theoretical and biological arguments strongly suggest that building such
systems requires models with deep hierarchical structure that support
inferences at multiple levels.
In this talk, I will introduce a broad class of probabilistic generative
models called Deep Boltzmann Machines (DBMs), and a new algorithm for
learning these models that uses variational methods and Markov chain Monte
Carlo. I will show that DBMs can learn useful hierarchical representations
from large volumes of high-dimensional data, and that they can be
successfully applied in many domains, including information retrieval,
object recognition, and nonlinear dimensionality reduction. I will then
describe a new class of more complex probabilistic graphical models that
combine Deep Boltzmann Machines with structured hierarchical Bayesian
models, called Hierarchical-Deep (HD) Models. I will show how these models
can learn a deep hierarchical structure for sharing knowledge across
hundreds of visual categories, which allows accurate learning of novel
visual concepts from few examples.
Biography
Ruslan Salakhutdinov received his PhD in computer science from the
University of Toronto in 2009. After spending two post-doctoral years at
the Massachusetts Institute of Technology Artificial Intelligence Lab, he
joined the University of Toronto as an Assistant Professor in the
Departments of Statistics and Computer Science. Dr. Salakhutdinov's
primary interests lie in statistical machine learning, computational
statistics, probabilistic graphical models, and large-scale optimization.
He is the recipient of the NSERC Postdoctoral Fellowship, Canada Graduate
Scholarship, and a Scholar of the Canadian Institute for Advanced
Research.
|
Oleg Golubitsky
Google Inc.
Probabilistic Prediction in Bidding Strategies for Online Ad Auctions
The order of ads returned for a given search query is a key decision that affects the user's satisfaction from the ads, the advertiser's ROI, and the publisher's revenue. This order is determined by an online ad auction, in which the bids defined by the advertisers are combined with a number of probabilistic predictions. Depending on the bidding strategy, the latter may include the predicted click-through rate, bounce rate, conversion rate, etc. Choosing a particular bidding strategy allows the advertiser to show their ads in a variety of different auctions with the same bid, delegating the optimization of the ROI under the changing auction conditions to the search engine's machine learning systems. In this talk, we will consider several bidding strategies, discuss the corresponding probabilistic prediction models, and show how their accuracy can be analyzed with the mapreduce framework.
Biography
Oleg Golubitsky received a PhD in Mathematics from Moscow State University and a PhD in Computer Science from the University of New Brunswick. He was an assistant professor of computer science at UNB and a postdoctoral fellow at the University of Pisa, Queen's University and the University of Western Ontario. Dr. Golubitsky's publications are in the areas of computer algebra, differential algebra, mathematical handwriting recognition, and quantum computing. He competed in and coached for the World Finals of the ACM International Collegiate Programming Contest. Since 2009, Dr. Golubitsky is a software engineer at Google.
|
Paul McNicholas
University of Guelph
Mixture model-based learning: The state of the art
Mixture-model based learning is reviewed within the unsupervised, semi-supervised, and supervised paradigms. Several special cases are studied, including longitudinal data, data of mixed type, non-Gaussian data, and microarray data. In each case, current and ongoing work is reviewed and illustrated with examples. Some implementation issues are discussed, along with proposed solutions. The lecture concludes with a summary and suggestions for future work.
Biography
Paul McNicholas was educated at Trinity College Dublin, Ireland, where he read mathematics (BA, MA), high-performance computing (MSc), and statistics (PhD). Paul's primary research concerns model-based clustering and classification. In addition, he has conducted research in other applications of mixture models, such as cure rates in survival analysis, and in data mining and sensometrics. Paul is currently the University Research Chair in Computational Statistics at the University of Guelph, where he is Associate Professor and Associate Chair at the Department of Mathematics & Statistics. He leads a large research group, which currently comprises over 20 highly qualified researchers. His group enjoys funding support from industry plus a variety of federal and provincial sources. Paul recently received an Early Researcher Award from the Ontario Ministry of Research and Innovation. Applications of Paul's research can be seen most readily in areas like bioinformatics, food authenticity and sensory science.
|
