2013 IPI Conference & Expo; Fort Laderdale, Florida, U.S.A.
Plenary presentation, Dan Mitchel, Peer Ghent and Onno Zoeter : The LA ExpressPark project: design, implementation, and initial results
The University of Edinburgh, Institute for Adaptive and Neural Computation, Edinburgh, UK
May 7th, 2013 Guillaume Bouchard , Machine Learning for Services (MLS) research group, will give a seminar: Convex methods for multi-view learning, link prediction and collective matrix factorization "Many applications involve multiple interlinked data sources, but existing approach to handle them are often based on latent factor models which are difficult to learn. At the same time, recent advances in convex analysis, mainly based on the nuclear norm (relaxation of the matrix rank) and sparse structured approximations, have shown great theoretical and practical performances to handle very large matrix factorization problems with non-Gaussian noise and missing data. In this talk, we will show how multiple matrices can be jointly factorized using a convex formulation of the problem."
The Machine Learning for Services (MLS) group at XRCE provides Xerox Corporation with state-of-the-art machine learning and computational statistics expertise.
The end of the 20th century saw the widespread use of information technology as a result of the surge of digital communication and computers. Now, our modern societies are about to experience a data processing revolution. The wealth of structured and unstructured data that is currently being generated by humans (and machines) will soon reach levels beyond our comprehension. This creates both new opportunities and challenges.
One emerging grand challenge is to enable machines to reason under uncertainty. By machine reasoning is understood the ability of computers to draw conclusions or propose solutions to an unseen problem, based on past noisy observations (i.e., experience). A less ambitious goal, and pre-requisite to machine reasoning, is to enable computers to convert information extracted from the deluge of data into knowledge readily comprehensible and actionable by non-experts. In order to succeed, it will be required to device specialized data models and to develop scalable learning and inference algorithms that can automatically organize, filter, aggregate, and summarize petabytes of distributed, possibly heterogeneous data. Once these new functionalities are integrated into commercial products and services, they will drastically change the way individuals, groups of individuals and organizations communicate, work, make decisions, or even innovate.
Our mission is to facilitate the adoption of data-intensive algorithms for automated decision making, the emergence of complex adaptive systems, and the widespread use of predictive analytics capabilities in organizations. Our statistical models capture individual preferences, enable self-learning services and/or borrow from behavioural economics in order to be incentive compatible. The resulting evidence-driven solutions are expected to play a key role in Xerox’ next generation document and business process outsourcing services .
Our current research activities are articulated along three main research lines:
Our research aims to advance machine learning methodology, including generative and discriminative data modelling, structured prediction, collaborative filtering, choice modelling, and relational learning. We pay special attention to principled ways to deal with uncertainty (noise and partial information) and large-scale data intensive processes. Our work requires strong expertise in Bayesian inference, as well as convex and nonlinear optimization. Our data-driven solutions are applied in a wide range of domains, including customer care (white paper ), healthcare (ITFoM: video ), transportation (LA Express: video ), participatory governance (FUPOL: www.fupol.eu ), and specialised search (FusePool: www.fusepool.eu ).
The team participates to the European network of Excellence PASCAL and is an associate partner of the European flagship IT Future of Medicine . Present research collaborations include co-supervision of one PhD student with Francis Bach at Ecole Normale Supérieure and two PhD students with Mark Girolami at University College London. The team also supports research projects funded by the Xerox Foundation with Frank Wood at Columbia University, Bin Yu at University of California Berkeley and Y. Narahari at the Indian Institute of Science.
If you are interested in joining the group for an internship, a PhD or a Post-doc, please contact Cedric Archambeau (firstname.lastname@example.org ).