Artificial Immune Systems: A New Computational Intelligence Approach
||Author: Leandro N. de Castro, Jonathan Timmis|
List Price: $54.95
Our Price: Click to see the latest and low price
Publisher: Springer Verlag (04 November, 2002)
Sales Rank: 396,282
Average Customer Rating: 3 out of 5
Customer ReviewsRating: 3 out of 5
Pretty good overview
Bio-inspired computing has taken the world by storm in the last few decades, going by the names of neural networks, genetic algorithms, evolutionary programming, and swarm intelligence. Another one has arisen has appeared in the last 15 years or so, is inspired by the biology of the immune system, and is the subject of this book. The authors of the book are aware that the approach is novel, but do a good job of presenting the field to newcomers (like myself), who want to know what it is all about and if it indeed has useful applications. They discuss their own work in the area and that of others, and extensive references are given for further reading.
After a short introduction to the subject in chapter 1, the authors move on to a description of the biological immune system in chapter 2. They stress the need for understanding the mechanisms that regulate the adaptive immune response, so as to be able to control the transformation of an immune response from an "aggressive" to a "benign" state. The authors explain the difference between the "innate" immune system and the "adaptive" immune system. As the name implies, the adaptive immune response is a kind of "learning" ability that allows the immune system to improve itself as antigens are encountered. The innate immune response though remains constant along the lifetime of the organism. A short description of the T-cells and B-cells is given, some of which can differentiate into "memory cells" that remain circulating in the body and protect against a given antigen. Particularly interesting is the role of "pattern recognition receptors" that recognize molecular patterns associated with pathogens. The clonal selection theory of the adaptive immune system, along with the somewhat controversial immune network theory.
Chapter 3 is an overview of how to to actually create an artificial immune system (AIS). The emphasize that anything deemed controversial in the biological framework need not be when viewed from a computational perspective, such as the immune network theory. Biology is used for the inspiration of the computational models, and as such they need not reflect entirely what is true in the biological case. They also emphasize that the various attempts to simulate the immune system on computers are not examples of an AIS. Also, an AIS is more than just a pattern recognition algorithm, even though it must employ this in its use. To give a framework for an AIS, the authors employ a model of immune cells and molecules called a "shape-space". In this shape space one models the affinity of the "molecules" via a metric, which the authors eventually choose to be the Hamming metric. They then give an overview of various algorithms for modeling the immune system, such as bone marrow, thymus, and immune network models, in addition to clonal selection algorithms. For those readers familiar with dynamical systems, the immune network models are very interesting, due to the use of differential equations, and also the fact that such in immune network models the immune system is performing even in the absence of external stimuli.
Chapter 4 gives a survey of artificial immune systems, such as spectra recognition for chemical reactions, infectious disease surveillance, analysis of medical data, and computational security. The latter was of particular importance to me, so I read the discussion and the references with more attention than other parts of the book. The issue with the approaches for network intrusion detection and virus detection lie mostly in the performance of the network. Agents that are cleverly designed may form a very accurate way of detecting this malicious behavior, but their deployment on a network may degrade the its performance considerably.
I did not read chapters 5 and 6 so I will omit their review.
In chapter 7, the authors discuss various case studies in artificial immune systems that shed more light on the examples of Chapter 4. The computer network security application is discussed again, and a low number of false positives is shown to follow after the artificial immune system is simulated. However, the performance of the network is not pointed out by the authors. The authors also give more details on the application of artificial immune systems to data analysis and optimization. The discussion is interesting, but it is still an open question as to whether this approach is indeed better than other ones in optimization theory, i.e. how does the immune approach compare with the "free-lunch" theorems so often quoted in optimization theory? The authors do make a brief comparison of their optimization algorithm with evolution strategies, and this is somewhat helpful to those who are familiar with the latter.
The last chapter of the book looks to future applications of artificial immune systems, and in its connection with learning paradigms in artificial intelligence. The authors are open-minded about the future of AIS but also subject it to critical analysis.
The book motivated me to investigate the use of AIS more fully, and to begin thinking about possible applications, such as 1. Event correlation in networks. 2. Network routing: Routes that are inefficient are viewed as "antigens", and the network immune system will then cure the system of these routes, meaning that it will remember them as being antigens up to some practical time scale. The routing scheme in place will not implement these routes within this time frame. 3. The TCP/IP protocol in the context of the immune network theory where reliable connections are based on the epitope/paratope recognition capability. Any emergent properties of the network overlaid with the TCP/IP protocol such as learning, memory, and self-tolerance could be studied by viewing the packet network as an immune network. 4. Network QoS, with packets marked as low priority viewed as temporary antigens. 5. Using the function optimization capabilities of AIS do calculate the effective bandwidth of ATM networks. 6. Data analysis, particularly in the construction of algorithms to distinguish chaos from noise.
· An Introduction to Genetic Algorithms (Complex Adaptive Systems)
· Swarm Intelligence: From Natural to Artificial Systems (Santa Fe Institute Studies on the Sciences of Complexity)
· Swarm Intelligence
· Applied Evolutionary Algorithms in Java
· Understanding 99% of Artificial Neural Networks : Introduction & Tricks