基本信息
- 原书名:Swarm Intelligence
- 原出版社: Morgan Kaufmann
- 作者: (美)James Kennedy Russell C.Eberhart Yuhui Shi
- 丛书名: 图灵计算机科学
- 出版社:人民邮电出版社
- ISBN:9787115195500
- 上架时间:2009-1-20
- 出版日期:2009 年2月
- 开本:16开
- 页码:512
- 版次:1-1
- 所属分类:计算机 > 人工智能 > 综合
编辑推荐
本书综合运用认知科学、社会心理学、人工智能和演化计算等学科知识,提供了一些非常有价值的新见解,并将这些见解加以应用,以解决困难的工程问题。
内容简介
目录
chapter one Models and Concepts of Life and Intelligence 3
The Mechanics of Life and Thought 4
Stochastic Adaptation: Is Anything Ever Really Random? 9
The “Two Great Stochastic Systems” 12
The Game of Life: Emergence in Complex Systems 16
The Game of Life 17
Emergence 18
Cellular Automata and the Edge of Chaos 20
Artificial Life in Computer Programs 26
Intelligence: Good Minds in People and Machines 30
Intelligence in People: The Boring Criterion 30
Intelligence in Machines: The Turing Criterion 32
chapter two Symbols, Connections, and Optimization by Trial and Error 35
Symbols in Trees and Networks 36
Problem Solving and Optimization 48
A Super-Simple Optimization Problem 49
Three Spaces of Optimization 51
Fitness Landscapes 52
High-Dimensional Cognitive Space and Word Meanings 55
前言
In this book we argue that what we do right is related to our sociality.We will investigate that elusive quality known as intelligence, which is considered first of all a trait of humans and second as something that might be created in a computer, and our conclusion will be that what-ever this "intelligence" is, it arises from interactions among individuals.We humans are the most social of animals: we live together in families,tribes, cities, nations, behaving and thinking according to the rules and norms of our communities, adopting the customs of our fellows, includ-ing the facts they believe and the explanations they 'use to tie those facts together. Even when we are alone, we think about other people, and even when we think about inanimate things, we think using language--the medium of interpersonal communication.
Almost as soon as the electronic computer was invented (or, we could point out, more than a century earlier, when Babbage's mechanical analytical engine was first conceived), philosophers and scientists began to ask questions about the similarities between computer programs and minds. Computers can process symbolic information', can derive conclu-sions from premises, can store information and recall it when it is appro-priate, and so on--all things that minds do. If minds can be intelligent,those thinkers reasoned, there was no reason that computers could not be. And thus was born the great experiment of artificial intelligence.
To the early Al researchers, the mark of intelligence was the ability to solve large problems quickly. A problem might have a huge number of possible solutions, most of which are not very good, some of which are passable, and a very few of which are the best. Given the huge number of possible ways to solve a problem, how would an intelligent computer program find the best choice, or at least a very good one? Al researchers thought up a number of clever methods for sorting through the possibili-ties, and shortcuts, called heuristics, to speed up the process. Since logical principles are universal, a logical method could be developed for one problem and used for another. For instance, it is not hard to see that strings of logical premises and conclusions are very similar to tours through cities. You can put facts together to draw conclusions in the same way that you can plan routes among a number of locations. Thus,programs that search a geographical map can be easily adapted to ex-plore deductive threads in other domains. By the mid-1950s, programs already existed that could prove mathematical theorems and solve prob-lems that were hard even for a human. The promise of these programs was staggering: if computers could be programmed to solve hard prob-lems on their own, then it should only be a short time until they were able to converse with us and perform all the functions that we the living found tiresome or uninteresting.
But it was quickly found that, while the computer could perform superhuman feats of calculation and memory, it was very poor--a com-plete failure--at the simple things. No AI program could recognize a face, for instance, or carry on a simple conversation. These "brilliant"machines weren't very good at solving problems having to do with real people and real business and things with moving parts. It seemed that no matter how many variables were added to the decision process, there was always something else. Systems didn't work the same when they were hot, or cold, or stressed, or dirty, or cranky, or in the light, or in the dark, or when two things went wrong at the same time. There was always something else.
The early Al researchers had made an important assumption, so fun-damental that it was never stated explicitly nor consciously acknowl-edged. They assumed that cognition is something inside an individual's head. An Al program was modeled on the vision of a single disconnected person, processing information inside his or her brain, turning the prob-lem this way and that, rationally and coolly. Indeed, this is the way we experience our own thinking, as if we hear private voices and see private visions. But this experience can lead us to overlook what should be our most noticeable quality as a species: our tendency to associate with one another, to socialize. If you want to model human intelligence, we argue here, then you should do it by modeling individuals in a social context,interacting with one another.
In this regard it will be made clear that we do not mean the kinds of interaction typically seen in multiagent systems, where autonomous subroutines perform specialized functions. Agent subroutines may pass information back and forth, but subroutines are not changed as a result of the interaction, as people are. In real social interaction, information is exchanged, but also something else, perhaps more important: individ-uals exchange rules, tips, and beliefs about how to process the informa-tion. Thus a social interaction typically results in a change in the think-ing processes--not just the contents--of the participants.
It is obvious that sexually reproducing animals must interact occa-sionally, at least, in order to make babies. It is equally obvious that most species interact far more often than that biological bottom line. Fish school, birds flock, bugs swarm--not just so they can mate, but for rea-sons extending above and beyond that. For instance, schools of fish have an advantage in escaping predators, as each individual fish can be a kind of lookout for the whole group. It is like having a thousand eyes. Herding animals also have an advantage in finding food: if one animal finds something to eat, the others will watch and follow. Social behavior helps individual species members adapt to their environment, especially by providing individuals with more information than their own senses can gather. You sniff the air and detect the scent of a predator; I, seeing you tense in anticipation, tense also, and grow suspicious. There are numer-ous other advantages as well that give social animals a survival advan-tage, to make social behavior the norm throughout the animal kingdom.
What is the relationship between adaptation and intelligence? Some writers have argued that in fact there is no difference, that intelligence is the ability to adapt (for instance, Fogel, 1995). We are not in a hurry to take on the fearsome task of battling this particular dragon at the mo-ment and will leave the topic for now, but not without asserting that there is a relationship between adaptability and intelligence, and noting that social behavior greatly increases the ability of organisms to adapt..
We argue here against the view, widely held in cognitive science, of the individual as an isolated information-processing entity. We wish to write computer programs that simulate societies of ind!viduals, each working on a problem and at the same time perceiving the problem-solving endeavors of its neighbors, and being influenced by those neigh-bors' successes. What would such programs look like?
In this book we explore ideas about intelligence arising in social con-texts. Sometimes we talk about people and other living--carbon-based--organisms, and at other times we talk about silicon-based entities, exist-ing in computer programs. To us, a mind is a mind, whether embodied in protoplasm or semiconductors, and intelligence is intelligence. The important thing is that minds arise from interaction with other minds.That is not to say that we will dismiss the question casually. The interest-ing relationship between human minds and simulated minds will keep us on our toes through much of the book; there is more to it than meets the eye.
In the title of this book, and throughout it, we use the word swarm to describe a certain family of social processes. In its common usage,"swarm" refers to a disorganized cluster of moving things, usually in-sects, moving irregularly, chaotically, somehow staying together even while all of them move in apparently random directions. This is a good visual image of what we talk about, though we won't try to convince you that gnats possess some little-known intelligence that we have discov-ered. As you will see, an insect swarm is a three-dimensional version of something that can take place in a space of many dimensions--a space of ideas, beliefs, attitudes, behaviors, and the other things that minds are concerned with, and in spaces of high-dimensional mathematical sys-tems like those computer scientists and engineers may be interested in.
We implement our swarms in computer programs. Sometimes the emphasis is on understanding intelligence and aspects of culture. Other times, we use our swarms for optimization, showing how to solve hard engineering problems. The social-science and computer-science ques-tions are so interrelated here that it seems they require the same answers.On the one hand, the psychologist wants to know, how do minds work and why do people act the way they do? On the other, the engineer wants to know, what kinds of programs can I write that will help me solve extremely difficult real-world problems? It seems to us that if you knew the answer to the first question, you would know the answer to the second one. The half-century's drive to make computers intelligent has been largely an endeavor in simulated thinking, trying to understand how people arrive at their answers, so that powerful electronic computa-tional devices can be programmed to do the hard work. But it seems re-searchers have not understood minds well enough to program one. In this volume we propose a view of mind, and we propose a way tO imple-ment that view in computer programs--programs that are able to solve very hard mathematical problems.
In The Computer and the Brain, John von Neumann (1958) wrote, "I suspect that a deeper mathematical study of the nervous system... will affect our understanding of the aspects of mathematics itself that are in-volved. In fact, it may alter the way in which we look on mathematics and logics proper." This is just one of the prescient von Neumann's pre-dictions that has turned out to be correct; the study of neural systems has opened up new perspectives for understanding complex systems of all sorts. In this volume we emphasize that neural systems of the intelligent kind are embedded in sociocultural systems of separate but connected nervous systems. Deeper computational studies of biological and Cul-tural phenomena are affecting our understanding of many aspects of computing itself and are altering the way in which we perceive comput-ing proper. We hope that this book is one step along the way toward that understanding and perception.
A Thumbnail Sketch of Particle Swarm Optimization
The field of evolutionary computation is often considered to comprise four major paradigms: genetic algorithms, evolutionary programming,evolution strategies, and genetic programming (Eberhart, Simpson, and Dobbins, 1996). (Genetic programming is.sometimes categorized as a subfield of genetic algorithms.) As is the case with these evolutionary computation paradigms, particle swarm optimization utilizes a "popula-tion'' of candidate solutions to evolve an optimal or near-optimal solu-tion to a problem. The degree of optimality is measured by a fitness func-tion defined by the user.
Particle swarm optimization, which has roots in artificial life and so-cial psychology as well as engineering and computer science, differs from evolutionary computation methods in that the population members,called particles, are flown through the problem hyperspace. When the population is initialized, in addition to the variables being given random values, they are stochastically assigned velocities. Each iteration, each particle's velocity is stochastically accelerated toward its previous best position (where it had its highest fitness value) and toward a neighbor-hood best position (the position of highest fitness by any particle in its neighborhood).
The particle swarms we will be describing are closely ;elated to cellular automata (CA), which are used for self-generating comp. uter graphics movies, simulating biological systems and physical phenomena, design-ing massively parallel computers, and most importantly for basic re-search into the characteristics of complex dynamic systems. According to mathematician Rudy Rucker, CAs have three main attributes: (1) indi-vidual cell updates are done in parallel, (2) each'new cell value depends only on the old values of the cell and its neighbors, and (3) all cells are updated using the same rules (Rucker, 1999). Individuals in a particle swarm population can be conceptualized as cells in a CA, whose states change in many dimensions simultaneously.
Particle swarm optimization is powerful, easy to understand, easy to implement, and computationally efficient. The central algorithm com-prises just two lines of computer code and is often at least an order of magnitude faster than other evolutionary algorithms on benchmark functions. It is extremely resistant to being trapped in local optima.
As an engineering methodology, particle swarm optimization has been applied to fields as diverse as electric/hybrid vehicle battery pack state of charge, human performance assessment, and human tremor di-agnosis. Particle swarm optimization also provides evidence for theo-retical perspectives on mind, consciousness, and intelligence. These theoretical views, in addition to the implications and applications for engineering and computer science, are discussed in this book.
媒体评论
——Genetic Programming and Evolvable Machines
“这本书极为出色,不愧为PSO和群体智能的最佳参考书。..
——Konstantinos E.Parsopoulos,希腊Patras大学...