The overall objective of the Working Group is to demonstrate the effectiveness of technologies that arise from a deep understanding of the performance and implementation of learning systems on real world data. As the robustness and efficacy of this technology increases the range of applications of the developed techniques will correspondingly multiply. The Working Group aims to specify novel systems and prototype them on real world data supplied by industrial collaborators or other benchmarks.
The Work will be divided into the following three Areas:
- Foundations and Algorithmics of Learning
Novel models and techniques will be developed for analysing learning systems. Models will be considered that more accurately reflect real world conditions. Algorithms will be designed which perform well in the developed models.
- Specification and Prototyping of Novel Learning Systems
Novel insights will be translated into specifications for practical learning systems, which take advantage of the fundamental understanding, achieved and hence either have improved performance or more efficient learning algorithms. Where appropriate the systems will be prototyped on industrial applications or benchmark data.
- Real Valued Complexity and Function Classes
The task of assessing the computational richness of function classes and their suitability for use in learning systems will be carried forward.
The funded activities of the group will be as follows:
- Workshops on particular themes of interest to a subset of the teams, including one per year focussed on industrial applications;
- Short visits by researchers from one site to another;
- Attendance at conferences and meetings related to the work;
- Yearly project meetings;
- Bringing external speakers to workshops or project meetings.
A number of companies will be associated with the Working Group and will be invited to attend the workshops and yearly meetings. In addition, the research community will benefit as insights gained in our research are made accessible via WWW, conferences and journals.