Advanced Interaction:
Agents
Investigators: Russell Beale, Bob Hendley, Peter Lonsdale, Justin Wilkes
Associated grants: Microsoft, Mobile Devices - EPSRC Capital Equipment
We have developed a coherent architecture that allows the rapid development and integration of intelligent and autonomous agents. Architectures for agent systems (cameo, aQtiveSpace) and intelligent agents (onCue) were developed here and commercialised through aQtive. There are numerous projects based on these principles.
Data mining
Russell Beale, Bob Hendley
Data Mining with the Haiku system is based on the premise that machine learning is excellent at processing large amounts of data, but often produces unintelligible or irrelevant results. On the other hand, humans have a fantastically well-adapted visual system which we can utilise to pick out gross areas of interest, but they are poor at discerning statistically-viable relationships between data items, especially if there are a large number of them. This project ties together the two systems, human and computer, to produce a much more effective data mining system.
The work on intelligent agents is now being coupled with previous work on adaptive semantic genetic algorithms, used to produce tailorable rules, and this is coupled to other work on autonomous self-organising high dimensional visualisations.
Visualisation
Bob Hendley
An application of the intelligent agents, we're also looking at visualisation for search, user understanding, and information display. The approach taken is to create interaction between elements of the visualisation according to some characteristics of the data, and allow the pictures to self-organise over time. We are also looking at growing organic structures to represent complex high-dimensional data.
Management not manipulation
Russell Beale
As devices get smaller and we get more and more agents doing our work for us, we need to move from direct manipulation to management: we need to understand what agents are doing on our behalf, and where, and how long they'll be. The project is investigating suitable approaches to this, investigating metaphors and appropriate tools.
Agent Object Management
Many things that we use are accessed by others - shared calanders, copies of papers, web pages, and so on. This project is developing a tool to monitor, observe and control such things. Certain web technologies allow us to be notified, but more complex management of who is interested in you is likely to prove much more useful.