Engineering Working Group

Andrew Tuson, City University, London

1) Overview

This session took a broad view of AIS optimisation in engineering, control and related areas such as scheduling. The group consisted of an industriali st (with interests in mobile robots) an academic engineer (with interests i n control of thermal processes), an academic computer scientist (scheduling and OR applications) and two PhD students (scheduling and software testing respectively).

The approach taken was to sort the keywords elicited during the poster sess ion into groups according to the interests of the attendees. An attempt was made to "tease out" concerns and barriers to AIS adoption in the areas of interest as well as identify "targets" where AIS could provide capability t hat is not currently available.

2) Themes

A number of themes for development arose from the discussion.

2.1) Why AIS? What Does it buy us?

Some concern was raised regarding ensuring that that mistakes made by, e.g. the EC community, are not repeated. This means not reinventing the wheel, and making it clear (and honest!) what AIS is especially useful for and wha t it is not.

It was noted that applications research is interdisciplinary by its very na ture, though in the case of AIS likely to be mostly with the "CS" side of A IS rather that the "Biology" side. Applications will (and should) be a majo r driver of AIS technique development.

2.2) Methodology and "Good Practice"

Practitioners will likely be domain experts and not necessarily AIS researc hers. They want something they can adapt, reconfigure and use!

Therefore the AIS community should keep in mind that practitioners will nee d guidance in structuring a diverse set of approaches and deciding what is relevant (or not).

A codified body of good practice would help and we should aspire to produci ng this.

2.3) Where AIS Might Win

There was definite interest in the robust, emergent regulatory nature of im mune systems. There appears to be a real need for these kinds of capabiliti es in control systems, e.g. in mobile robotics.

It was felt AIS would be unlikely to produce clear gains for conventional " static" scheduling applications (other than a case-by-case "suck it and see " approach). There were, however, two scheduling applications identified th at AIS potentially may have a clear impact. The first is in area of schedul e recovery/repair; this was inspired by Emma Hart's immune library work. Th e second is looking at viewing reactive scheduling/workflow management as a AIS control/regulation problem (in a similar way to the above).

3) Actions.

Two sets of actions arose.

3.1) Promoting Good Practice in Applications

In general any actions by ARTIST that provide resources and dissemination o f guidelines and good practice to potential practitioners would be welcome. Suggestions included: distilling key issues and approaches, guidelines and case studies of good practice, providing open source implementations of AI S and datasets of problems relevant to practitioners (so AIS techniques are developed and evaluated on realistic problems).

At the reporting session, it was noted that ARTIST have already committed t o taking many of these actions, albeit at an early stage of development.

3.2) Two Research Themes, Two Retreats?

The discussion clearly identified two areas of potential AIS application wh ere capability building would likely be successful (and appreciated).

- Self Regulation/Holistic Control

- Reactive Scheduling and Schedule Recovery

The obvious next stage is to gather interested parties and hold a workshop/ retreat to make this happen.

Interested parties, please come forward!

Top...

Return to Workshop Page