Data Mining Working Group

1. Why use an AIS? Does an AIS bring real, tangible benefit (over and above a traditional technique, or a GA) or a novel viewpoint?
2. What sort of collaboration makes sense? (for data mining, maybe the computer scientist acts as a bridge between a biologist and a professional data miner. Often we'll need a domain expert too).

Within data mining, we looked at both data manipulation and application areas formulate questions/proposals

• Data Manipulation
• Question
• Preprocessing
• Time series analysis
• Clustering

Why use AIS?

Normalisation, data cleaning, distance metrics, feature extraction/selection/representation are all issues general to data mining, and not specific to AIS.  Can learn from only 1 class ('normal' behaviour) - but there are issues with testing this!  AIS may be effective at clustering sparse data, and/or dealing with a dynamic environment

What sort of collaboration makes sense?

Collaboration (with data mining engineers/domain experts) is important, but the activity is likely to be part of an ARTIST project, not the main thrust.  Domain experts, especially to help with test data Modellers, who can shed light on the network dynamics of the biological system?

What proposals are there?

That said, even here there is an interesting possibility - the use of the innate immune system metaphor for data cleaning and/or initial filtering. Perhaps also a way of incorporating domain knowledge? If the domain changes, can we incorporate evolution of the innate system. Algorithm is done & dusted (or is it?) so novel application domain (we've had milling data, cooling units...) Can the AIS clustering algorithms be improved through biological insights? Not Jerne's idiotypic network incidentally, which is out of favour.

• Application Areas
• Question
• Danger
• Dynamic Environments

Why use AIS?

This is a uniquely immunological concept, though interesting possibilities arise through thinking about danger, not as bad/damage but as good, interesting or simply context.  AIS is (arguably) good for changing environments

What sort of collaboration makes sense?

Biological collaboration is probably helpful here. We need to pin down the metaphors (what is a tissue? APC? danger signal? antigen? antibody?) and the biologist needs to explain the danger pathway. By building a model that shows how these elements interact, we could even feed back to biology - ie a type IV project! Possibility for truly collaborative project (see below) as well as straight application work.

What proposals are there?

Possibilities abound (some pre-exisiting projects). eg

• Andy Secker's email filtering,
• Personlisation (danger = interest)
• Ubiquitous computing (danger=context)
• Building a generic danger model How is the environment changing? (speed, degree...) and how well does it generalise?

.. leading to an idea which might have benefits for both CS applications and biological understanding...

How does AIS cope with different types of change and how can we improve it? Within data mining, we looked at both data manipulation and application areas formulate questions/proposals

• Data Manipulation
• Question
• Preprocessing
• Time series analysis
• Clustering

Why use AIS?

Normalisation, data cleaning, distance metrics, feature extraction/selection/representation are all issues general to data mining, and not specific to AIS.  Can learn from only 1 class ('normal' behaviour) - but there are issues with testing this!  AIS may be effective at clustering sparse data, and/or dealing with a dynamic environment

What sort of collaboration makes sense?

Collaboration (with data mining engineers/domain experts) is important, but the activity is likely to be part of an ARTIST project, not the main thrust.  Domain experts, especially to help with test data Modellers, who can shed light on the network dynamics of the biological system?

What proposals are there?

That said, even here there is an interesting possibility - the use of the innate immune system metaphor for data cleaning and/or initial filtering. Perhaps also a way of incorporating domain knowledge? If the domain changes, can we incorporate evolution of the innate system. Algorithm is done & dusted (or is it?) so novel application domain (we've had milling data, cooling units...) Can the AIS clustering algorithms be improved through biological insights? Not Jerne's idiotypic network incidentally, which is out of favour.

.. leading to an idea which might have benefits for both CS applications and biological understanding...

How does AIS cope with different types of change and how can we improve it? Supplementary notes on data mining (to be put on ICARIS web site)

Top...

Return to Workshop Page