Rating: 
Summary: Needs more details, but a good introduction.
Review: The authors of this book state therein that "mind is not found in covert, private chambers hidden away inside the individual, but exists out in the open; it is a public phenomenon." This would be a very difficult claim to prove from a scientific standpoint, requiring an understanding of neuroscience, consciousness, and psychology that is not yet available. The author's intent though is more modest: they want to use this statement, which they encapsulate as "swarm intelligence", as a guide to finding successful optimization algorithms. They spend many pages discussing the foundations and background behind their approach, perhaps in too much detail given the usual pragmatism exhibited by many who study algorithms. Swarm intelligence is a relatively new paradigm in the field of optimization, but its justification should come from the results it gives in practical optimization problems, not in the broad philosophical language that predominates the first part of the book.
Particle swarm optimization is introduced in the book in both 'binary' and 'real-valued' form. The authors identify three principles behind the workings of particle swarms, namely the tendency to "evaluate"; the use of comparisons to others as a way of measuring individual status or progress; and the use of imitation. These three principles they say allow individuals to adapt to highly complex environments and solve very difficult problems. A binary decision model is used to introduce binary swarm algorithm, which is given in pseudocode, and is tested using a binary-coded version of the De Jong suite of test problems for optimization algorithms. A particle swarm model over the real numbers is then discussed, along with pseudocode, Both the binary and real models of particle swarms illustrate the fact that particle swarm optimization is a consequence of social interaction. The particles or "individuals" in the swarm learn from each other, and move to become more similar to their neighbors based on the knowledge obtained. Particle swarm optimization is dependent on the existence of social structure, the latter of which is determined by the formation of neighborhoods. These neighborhoods can have a different topology, determined solely by the numerical indices assigned to each individual.
The pseudocode given for particle swarm optimization illustrates well the basic workings of the algorithm in terms of the "local" and "global" viewpoint of the particles in the swarm. First the swarm is initialized and the performance of each particle is evaluated using its current position. The performance of each individual is then compared to its best performance so far, and the velocity for each particle changed according to a formula dependent on a system parameter. Each particle is then moved to a new position and the entire process repeated until convergence is attained. When a particle is very far from its best solution previously found, the change in velocity will be greater in order to return the particle toward its best solution. The system parameter will govern how much the particle trajectories oscillate, with smaller values of this parameter ensuring smoother trajectories. The authors give examples with graphs to illustrate this behavior and the influence of the system parameter.
Being aware that particle swarm optimization is typically viewed as a kind of evolutionary algorithm, the author address in some detail the reasons for this classification and its justification. Acknowledging that particle swarm algorithms have been influenced by evolutionary computation, they discuss some of the differences between the two approaches. In evolutionary algorithms individuals survive according to their fitness, whereas in particle swarms every individual will survive. In addition, in particle swarms, it is the velocities that are adjusted, whereas in evolutionary computing it is the positions that are state. The authors express this by saying that it is the "fate" rather than the "state" that is altered in particle swarm optimization.
The authors include an entire chapter on applications in the book, one of them being the use of particle swarms to evolve neural networks. Evolved neural networks have been shown to perform better in some cases than ones designed from scratch. After discussing some of the approaches to evolving neural networks, the authors point out, correctly, that hardly any of the studies in evolving neural networks are quantitative studies of how well they perform relative to other approaches Performance metrics are hardly ever given, which would allow interested parties to make objective and intelligent decisions on which approach is the most viable. The author's approach of using particle swarms to evolve neural networks also, interestingly, involves evolving the transfer functions of the neural networks, and they test their approach by using the Iris Data Set, a frequent benchmark for classification algorithms. Preliminary results indicate that their approach is a viable one and that it shows promise, but they admit that further experiments are needed in order to form valid conclusions.
So are the optimization algorithms based on swarm intelligence better than those that are based on, for example, on evolutionary algorithms? Are they better than those that are purely randomized algorithms? The authors are not shy about discussing how swarm intelligence optimization algorithms compare with other optimization algorithms, particularly randomized algorithms and the now famous "free-lunch" theorems of David Wolpert and William Macready. They discuss the free-lunch theorems via a very interesting example dealing with finding one's way out of a room. Using this example, they are convincing in their claim that even though no algorithm can be said to be better than any other when averaged over all cost functions, this averaging is done over processes or tasks that might be deemed absurd in the context of many problems of practical interest. Thus for "real" problems, one algorithm might indeed be "better" than another.
Rating: 
Summary: Intelligence is social
Review: Besides its excellent presentation and highly methodological structure, this book describes the anti-individualistic point of view of "intelligence". The point is about showing how problem solving and evolution can be thought as mainly driven by the way components interact, not through some notion of an individual's "intelligence" that would be the property of the person (or any "entity") which is solving a problem, and not the result of the contacts with his (its) environment.If you like books providing a different point of view of AI this is the book you are looking for...in addition it is entertaining and easily readable.
Rating:
Summary: Corollary Information Also Great
Review: I concur with the other postive reviews of this book as a very interesting application of multiple disciplines to the field of evolutionary computing and would also add that it provides some very palatable explanations of other concepts like classic genetic algorithms, including supporting maths (which, as someone without a terribly strong background in math, I really appreciated) as well as some good heuristics for use in real systems. Where other books might treat you to a cursory introduction and then start speaking to you as if you were a research peer (which *can* stimulate the reader to higher knowledge, but often alienates as well), this one ramps up to it's more interesting ideas, ensuring you are not left to have to research elsewhere simply to understand it's contents and implications. Even if you are experienced in any of the fields covered by this book, I doubt the introductory material will bore you (or it's easy enough to skim). Certainly the challenging conclusions will not! Good work, Mr. Kennedy!
Rating:
Summary: Interesting Paradigm
Review: It's an immersive and powerful piece of scientific metrics and theoretical paradigm presentation. It shows that life can be a much deeper form of existence. The book presents the complexities of PSO in its network relativity but can be created using simple algorithms. The basis comes from the behavioural science andsocial patterns of insects such as bees and ants. Their process of colonial interaction and food foraging can be applied as a strong mathematical structure to computational science, robotics, and network technology. At the same time, you can take the exact principles -- in its raw idea -- and apply it to economic structure and business dynamics. I love how this book harks back to the parable of the blind men trying to explain what an elephant is like.
Rating:
Summary: Mind is Social
Review: My original motivation for reading Swarm Intelligence was a desire to learn about the Particle Swarm Optimization (PSO) algorithm -- in particular, to learn how to implement it in a Java program. To the credit of its authors, what I found in Swarm Intelligence was far more than that. The authors have taken on the rather daunting task of presenting a new paradigm -- a new way of thinking about mind and intelligence -- and they have succeeded. PSO, itself, is deceptively simple. The heart of the algorithm can be written in a single line of code. Understanding the basis for its approach to intelligence isn't difficult, either. The authors begin their explanation using the old parable about the blind men and the elephant. You are most likely familiar with the story. In summary form, it is about a group of blind men standing around an elephant each declaring "what an elephant is like" based upon which part of the elephant they are touching -- and elephant is like: a wall (side); a tree trunk (leg); a hose (trunk); a fan (ear); and so on. What is wrong with this story, the authors point out, is its implicit assumption that these blind men are also deaf. If not, as they each announced their impressions the individuals, as a group, would discover much more about what an elephant is. The significance here is easily missed. The capabilities of a group emerge from the individuals immersed in it. The group can do more (see more, discover more, experiment more) than the individuals from which it emerges and, by virtue of their immersion in it, the individuals benefit (and in turn, the group then benefits as it now emerges from these "benefited" individuals). The authors view this emergent/immergent "cycle" as the driving force behind mind and intelligence. In contrast to the normal (phenomenological) view of mind as an internal, private "thing that thinks," the authors assert that mind is something requiring sociality. To put it bluntly (and the authors do), in the absence of social immersion there is no mind; mind is social. The majority of the book is focused on this: why it's true, how it's true and how it is implemented in the PSO algorithm. It is easy to see how the book might have ended up a long philosophical argument. It isn't. Instead, the authors present a nicely written history of efforts to achieve "computational intelligence" (a much better phrase than the more familiar "artificial intelligence") including great summaries of evolutionary approaches, fuzzy logic, neural nets and artificial life. Along the way they point out recent advances in psychology and sociology. The net effect is that they don't need to argue their point. By the end of this part of the book the importance of sociality has become rather obvious. If you are interested in sociology, psychology, engineering and/or computer science you will enjoy this part of the book immensely, learn a lot and find a wealth of references to additional sources of information. The second part of the book presents the PSO algorithm, compares its performance with other methodologies (in addition to being simpler to understand and implement, it's an order of magnitude faster when applied to certain problems -- training neural nets, for example), demonstrates how it is applied to some "real life" problems and discusses some implications of (and speculations about) the approach. As with the first part of the book, the presentation is clear, concise and informative. There is, though, indications here that the PSO approach is rather new (young). There isn't enough experience with PSO yet to give this part of the book the same feeling of depth one gets from the first part. It's worth noting that the presentation (and description) of the PSO algorithm is done in mathematical terms. I would have much preferred a programming approach (using pseudo code) not because the math is too difficult (it's not) but because I haven't been "immersed in a mathematically minded social group" for many years. The almost exclusive use of Greek letters for symbols (variables) made reading difficult. Not only are they visually unfamiliar, I don't know their pronunciations (to illustrate the difficulty by way of analogy, consider the difference between reading "y equals b times x plus z" and "xgt equals kqj times yxf plus ktv"). I ended up rewriting the formulas in more familiar terms (using the text to figure out what the symbols represent when necessary) before I felt that I understood them. Mentioning my problem with the math is not meant to criticize but to suggest that the book could have been made accessible to more people had it also contained a more readable (and retainable) form of the algorithm, perhaps in an appendix. A good analogy of the PSO approach (more detailed than the "blind men" story) would also have been helpful. The only real criticism I have of the book's content is a minor one. Being as it is focused on the social requirements for mind, it tends to overlook the degree of individuality required to make PSO work. The algorithm, itself, has variables which control the expression of individuality and without which it could not work (at least not well), but this flipside to the social nature of the algorithm is never discussed as such. PSO works well precisely because it maintains the rather chaotic balance between the effects of sociality and individuality. The book presents a rather one-sided view of this balance. An aside for programmers: There is a companion site (of sorts) on the web for the book through which you can download Visual Basic and C source code of PSO implementations. There is also a Java applet which demonstrates PSO applied to a number of test functions but the source code for it is not available. There will also be an open source Java implementation as soon as I can make one available.
Rating:
Summary: Very good.. Thought provoking
Review: Really interesting visions are described in this book
Rating:
Summary: Good, but could have been more concise.
Review: Swarm intelligence is burdened with an awful lot of material that is not core to PSO. A great deal of the book consists of the philosophical ramblings of the authors, rather than technical treatment of the topic at hand. An even larger chunk of the book was devoted to what was essentially a survey of AI: neural nets, evolutionary programming, heuristics, etc. Much too much space was devoted to grounding the reader in AI before proceeding. I must admit, however, that, while I found it out of place, the 'AI primer' part of the book is one of the most useful and lucid I have seen; I just think that it should have been a separate book (and this one should have been much thinner). The material that is specific to PSO is a very small fraction of the book, but is thorough and accessible; there really are few alternatives if one is particularly interested in PSO. However, if you are just interested in emergent behavior, and its applications to AI, take a look at Ant Colony Optimization (Dorigo). It covers ACO, rather than PSO, but is more more readable, and provides a much better technical treatment of the topic, if you want to avoid the philosophy and primer.
Rating:
Summary: A good, readable survey of PSO techniques
Review: The book contains: a) An overview of evolutionary programming techniques. b) An exposition of the argument that intelligent behavior has a large social component in addition to a genetically determined component. c) The presentation of an optimisation technique whereby a swarm of possible solutions fly through a problem space and base their search trajectories not only on personal experience but also on the experiences of the group. ie- There is a social component to the search of the problem space. The presentation of (a) and (b) was quite good and readable. The presentation of (c) I found to be a little bit unclear. The algorithm is quite simple, and can be expressed succinctly, but I ended up having to go to secondary sources (web site and PSO C code) to understand exactly what they were doing. The title of the book seems to suggest the swarm develops an emergent property of intelligence. This is over-reach, and is probably not an interpretation that the authors would place on the PSO algorithm. The PSO algorithm is an interesting numeric optimisation technique, and it seems to be a more organic approach to developing neural network weights than techniques like back-propagation of errors. Overall, a good book that I would recommend. Points off for not being clearer in explaining the algorithm details.
Rating:
Summary: Misleadingly Fun
Review: The concepts of intelligence and thought have been the source of speculation and wonder since the dawn of mankind (or is it personkind?). With the advent of modern computers, computational systems were developed that were capable of some degree of artificial intelligence. However, the conceptual frameworks were difficult to understand and were even harder to implement. In this book, the authors lead us through a wonderful journey of the foundations of our thoughts, intelligence and psychology all the while taking us on a tour of the new field of evolutionary computing and its newest member - Swarm Intelligence. The authors begin with an excellent overview of the text that helps set the tone for the reader. This is probably one of the few times where reading the introduction actually enhances the enjoyment of the book. In the first several chapters, we are introduced to models and concepts of life, intelligent thought and computational intelligence. In so doing, great care has been taken to represent the diverse and divergent opinions on these subjects. The second section of the book is dedicated to explaining the concepts involved with the particle swarm and collective intelligence. Included in this section is a discussion of the partical swarm in relation to other techniques of evolutionary computing. Several "real-world" applications have been included and help clarify the utility of particle swarm in evolutionary computing. Overall, the book is well written, comprehensive and fun for anyone interested in intelligence or evolutionary computing. The variety of viewpoints only serves to make the book more engaging and superb reading; even for those who have little programming background. I HIGHLY RECOMMEND IT!!
Rating:
Summary: The best reference on PSO and Collective Intelligence
Review: This book is fantastic!
It consists of two parts. In the first part, the main ideas behind Evolutionary Computation and social behavior are tangibly described. A brief review of the most known evolutionary computation algorithms is provided and social behavior modeling issues are reported to prepare the reader for the second part. The second part is devoted to the Particle Swarm Optimization (PSO) algorithm and its applications. Both binary and real variants of PSO are considered and several theoretical aspects are investigated. The book closes reporting several applications and insightful conclusions. Perhaps the best book on collective intelligence and PSO.
|
