Hassan Masum

Statistical analysis of heuristics for evolving sorting networks

Designing efficient sorting networks has been a challenging combinatorial optimization problem since the early 1960s. The application of evolutionary computing to this problem has yielded human-competitive results in recent years. We build on previous work by presenting a genetic algorithm whose parameters and heuristics are tuned on a small instance of the problem, and then scaled up to larger instances. Also presented are positive and negative results regarding the efficacy of several domain-specific heuristics.

Review of a new kind of science

A New Kind of Science uses a wide range of easy-to-understand models mostly cellular automata variants to explore one key idea: some simple computational models can generate “very complex” behavior, and may describe aspects of the physical and mathematical universe better than traditional mathematical approaches. Wolfram’s book has been the subject of much speculation, partly due to its lengthy preparation time (10 years). Since it came out a few months ago, published opinions of the book have ranged from “worthless” to “ground-breaking.” As you will glean from points made later, this book is a difficult one to review objectively, for two main reasons: i) the writing style, and ii) the sheer breadth of questions raised. What follows is therefore a personal description of how the book did and did not impress me. On the positive side, Wolfram raises many interesting questions. He specifies a remarkably diverse range of models related to cellular automata variants. And he explains his ideas simply, with beautiful pictures and a main text which is accessible to any reader with a high-school education who is willing to think a bit. On the negative side, many of the ideas in the book have been discussed elsewhere, so there is not much material that is fundamentally new (or immediately usable). Some of the reasoning (such as the lack of differentiation between degrees of complexity) seems incorrect. And the immodest writing style and relative lack of acknowledgment and credit to other researchers is distasteful. While it asks many interesting questions in a clear if sometimes overstated way, this book does not create “a new kind of science.” It does, however, bring together many key issues and simple models for the interested reader and may well stimulate research into the questions and models proposed.

Switching models for nonstationary random environments

Learning automata are stochastic finite state machines that attempt to learn the characteristic of an unknown random environment with which they interact. The fundamental problem is that of learning, through feedback, the action which has the highest probability of being rewarded by the environment. The problem of designing automata for stationary environments has been extensively studied. When the environment is nonstationary, the question of modeling the nonstationarity is, in itself, a very interesting problem. In this paper, the authors generalize the model used in Tsetlin (1971, 1973) to present three models of nonstationarity. In the first two cases, the nonstationarity is modeled by a homogeneous Markov chain governing the way in which the characteristics change. The final model considers the more general case when the transition matrix of this chain itself changes with time in a geometric manner. In each case the authors analyze the stochastic properties of the resultant switching environment. The question of analyzing the various learning machines when interacting with these environments introduces an entire new avenue of open research problems. >

Review of Evolutionary computation and bioinformatics by Gary B. Fogel and David W. Corne, Morgan Kaufmann publishers, Inc., 2002

C urrent methods of gene function analysis have been likened to firing a gun at random parts of a roomful of radios and then trying to figure out why each one does or doesn’t work. A typical method is to mutate (or “knock out”) genes and identify any resulting functional deficits. Evolutionary Computation in Bioinformatics exemplifies one line of research toward a more holistic approach. The editors have written the first two introductory chapters and an appendix of relevant Internet resources. The remaining 14 chapters are highquality offerings from a variety of contributors. Because many geneticists may not know about evolutionary computation (EC) (and vice versa), the first two chapters are an introduction to each field from the point of view of the other. The first chapter (Bioinformatics for Computer Scientists) is well written, concisely covering basic ideas such as protein structure and the genetic code. Corne and Fogel make the good points that sequencing is only an early milestone in the long road to deciphering organism function and that many traditional biological models oversimplify for reasons of tractability and wind up with “the right answer to the wrong problem” (e.g., protein folding approximations, ecological models). This is a primary motivation in the use of EC to make more realistic models tractable. A little more detail in this chapter would have been welcome, e.g., on regulatory networks and protein types and function, but some pointers are given to introductory biological literature. I found the second chapter (EC for biologists) to be rather weak. Despite its modest length, the reader is left with the impression of too much detail and not enough key ideas. The goal of this type of chapter should be to convey the essentials; instead, a number of uncommon model variants are mentioned, and there are few diagrams and insufficient computational detail to properly explain the key algorithms. The remaining chapters cover a wide variety of topics. On the whole, chapters are clearly written and of high quality, with many interesting techniques outlined. We’ll look at a sample to give the flavor of the book. Blazewicz and Kasprzak discuss sequence determination from experimental data. Given a fragment of DNA, how can one find its sequence? One approach is as follows: first, construct a microarray chip with all possible DNA fragments of some short length represented (this set of short DNA fragments is called the “oligonucleotide library”). Then make many copies of the original DNA fragment, attach a fluorescent marker to them, and introduce these to the microarray chip. As we know, DNA bonds to complementary strands, and hence the microarray chip will show those members of the oligonucleotide library that are attached to the DNA fragment, with those attached along their entire length fluorescing most brightly. The inferred set of oligonucleotides that make up the original DNA fragment is called the “spectrum”; the authors address the problem of deducing the original DNA fragment from its experimentally determined spectrum, where the spectrum determination is subject to errors. It’s as if one were to take this paragraph, chop it up into small bits of a few letters each, add a few errors, and EVOLUTIONARY COMPUTATION AND BIOINFORMATICS by Gary B. Fogel and David W. Corne, editors, Morgan Kaufmann Publishers, Inc., San Francisco, 2002 416 pp,

Review of Data Structures and Algorithms in Java (2nd ed) : Michael T Goodrich and Roberto Tamassia

65.95 (US hardcover) Reviews book & software

On modelling non-stationary random environments using switching techniques

Data Structures is a first book on algorithms and data structures, using an object- oriented approach. The target audience for the book is a second-year CS class introducing fundamental data structures and their associated algorithms. This second edition of the book has been corrected and revised, and is a substantial improvement over the first edition. A companion web site contains useful ancillary tools, including an excellent set of slides.Despite several minor errors and some questionable stylistic choices, I found this version of the book to be well-written. The problem sets are large, interesting, and thought-stimulating. In several places connections are drawn between the algorithm being discussed and important contemporary real-world problems (e.g. search engines, DNA sequence comparison, garbage collection), which students usually appreciate.

The turing ratio: metrics for open-ended tasks

Learning automata are stochastic finite state machines that attempt to learn the characteristic of a random environment with which they interact. The fundamental problem is that of learning, through feedback, the action which has the highest probability of being rewarded by the environment. The problem of designing automata for stationary environments has been extensively studied. When the environment is non-stationary, the question of modelling the non-stationarity is, in itself, a very interesting problem. In this paper, the authors generalize the model used in Tsetlin (1961, 1963) to present models of non-stationarity. In the first the non-stationarity is modelled by a homogeneous Markov chain governing the way in which the characteristics change. The final model considers the more general case when the transition matrix of this chain itself changes with time in a geometric manner. In each case the authors have analyzed the stochastic properties of the resultant switching environment. The question of analyzing various automata interacting with these environments is open.<<ETX>>