Objectives

Deep Computing is a term for methods solving complex and large-scale modeling and analysis problems with emerging computer systems that combine ultrafast processing with sophisticated analytical software. Deep Computing can be seen to consist of three intertwined research areas:

• Deep modeling: prediction and data mining with very large data sets,
• Deep optimization: computationally efficient optimization of complex multivariate cost functions, and
• Deep view: interfaces for understanding high-dimensional data.

The DeepC project has two major objectives:

• Methodological research objective. To develop the theory and methods required for obtaining very large-scale computational, data and communications capabilities that can be used to solve "grand challenge"-level Deep Computing problems in business and science. The research focuses on stochastic approaches and is methodological and theoretical in nature, and aims at topics that can have great impact in this area in the future.
• Applied research objective. To demonstrate solutions to previously intractable business and scientific problems by exploiting the advances in Deep Computing research in areas such as data modeling and analysis, high-end computing, search and optimization algorithms, and high-dimensional visualization. Such demonstrations will often be results of joint multi-disciplinary work together with scientists (scientific problems) as well as industry (business problems).

A prototype mobile user location system, developed by the CoSCo group. The location of a mobile terminal (red area) is inferred from a set of available signal strength measurements.		An example of a 3D visualization of a high-dimensional data set, produced by a novel Bayesian visualization technique developed by the CoSCo group. In this particular example, each point in the picture represents a session path of a WWW site visitor, and the colors of the points correspond to different type of user profiles.

Some of the methods developed in the project have been implemented in the B-Course software, a unique Bayesian web-based interactive tutorial and data analysis service publicly available at address http://b-course.cs.helsinki.fi.

Contact person

Professor Henry Tirri
Complex Systems Computation group (CoSCo)
Department of Computer Science
University of Helsinki, Finland