Artificial Immune Systems
Artificial Immune Systems
Artificial Immune System is the collective term for immune inspired computational methods. AIS's are abstractions of immunological concepts which have developed into a number of algorithms, applied to a range of problems. Such problems include computer security, bioinformatics, classification, clustering, fault tolerance and data mining. Many aspects of the immune system have been 'borrowed' in the field of artificial immune systems. This includes the negative selection process of T-cells in the thymus, antibody-antigen interaction and pattern matching as described by the idiotyipic network theory, the clonal expansion and responses of the B-cell population, memory cells for remembering solutions and recently the Danger theory.
When I was first introduced to the concept of an artificial immune system, my initial reaction was (predictably) "how can something that simple be classed as an immune system?" Of course, artificial immune systems simply borrow certain aspects- necessary to keep the complex interactions of the systems under wraps. However, I was still not that satisfied. Immunology is a really rich field full of amazingly complex interactions. I think that my major interest in AIS is involved in the development of new algorithms, which are more closely based on the immune system itself including the development of the Dendritic Cell Algorithm.
The book written by Jon Timmis and Leandro De Castro gives a good summary. For more technical details, then consult the proceedings of the International Conference On Artificial Immune Systems (ICARIS), or visit the AIS homepage at artificial-immune-systems.org.
The Dendritic Cell Algorithm
The Dendritic Cell Algorithm is an immune-inspired anomaly detection algorithm, which performs classification through correlation and filtering. It is inspired by an abstract model of the dendritic cells of the human immune system. These cells instruct the adaptive immune system how to respond appropriately to foreign invaders, namely pathogens, making them ideal inspiration for the detection of malicious behaviour in computer systems.
The DCA is derived from an abstract model of DC biology. The resultant algorithm is population based, with each cell in the population assigned a lifespan value, which is decremented upon the receipt of signal input. Different cells process signals acquired over different time periods, generating individual 'snapshots' of input information. When aggregated across the population, antigens are classified on the basis of the consensus opinion of whether a particular type of antigen is normal or anomalous. We believe this time window effect is responsible for the robust detection shown by the original DCA though have yet to prove that this is indeed the case. The robust detection and correlation performed by the DCA makes it a contender for the analysis of noisy, time-ordered data.
Metaphorically, natural DCs are the crime-scene investigators of the human immune system, traversing the tissue for evidence of damage - namely signals, and for potential suspects responsible for the damage, namely antigen. As with all things biological, it takes multiple DCs presenting multiple antigens to multiple effector T-cells for an actual response to be mounted. The combination of the population dynamics, signal processing and the correlation between signals and antigen make this system an effective and interesting metaphor for use within AIS.
Affective Computing
What gets your pulse racing? What scares the living daylights out of you? For me, I love a good rollercoaster and a freeride on my snowboard down a mountain, and I get scared when watching psychological thrillers. However, I doubt this is the same for everyone. How can we engineer technology using the detection of emotional responses to personalise the intensity of an experience. Adapting the interface to the user is an interesting and emerging area of computer science. The use of biosensor equipment and the analysis of biosensor data is a primary application area for my research. I am involved in understanding data collected from a number of large scale experiments, monitoring volunteers in a variety of situations, to see if it is possible to discriminate between thrill and fear using wearable biosensors. I find this problem particularly appealing as the data is vast, noisy and in real-time. I am currently working with horizon on an affective computing pilot project to construct emotional avatars.
