Associative Memory-Based Reasoning: A Computational Model of Analogy-Making in a Decentralized Multi-Agent Cognitive Architecture

From the Back Cover:

Throuhg analogy, novel situations and problems can be understood in terms of familiar ones. There is converging evidence that analogy-making lies at the very core of human cognition. Conversely, successful analogy-making requires the resources of an entire cognitive architecture. This book describes a model of analogy-making called AMBR (Associative Memory-Based Reasoning). AMBR is based on a hybrid symbolic-connectionist multi-agent cognitive architecture called DUAL. Macroscopic behavior in DUAL emerges from the interactions of simple processing agents in dynamic coalitions. Unlike the mainstream models of analogy-making, AMBR uses a decentralized representational scheme for problems and situations. The dynamic emergent processing of these decentralized representations is consistent with the context-sensitive and constructive nature of human memory. Both DUAL and AMBR were developed by Boicho N. Kokinov and his graduate students at New Bulgarian University. The book is a revised and expanded version of the author's Ph.D. thesis written under Prof. Kokinov's supervision at NBU. It will be of interest to cognitive modelers and cognitive scientists more generally.

Preface:

This book is based on my Ph.D. thesis, which was completed in July 1998 under the supervision of Dr. Boicho Kokinov at the New Bulgarian University (NBU) in Sofia, Bulgaria. The book describes a cognitive model of analogy-making developed in an effort to understand the mental processes that take place when a person perceives one situation as structurally similar to another. The model is called AMBR and is based on a hybrid symbolic-connectionist cognitive architecture called DUAL. Both DUAL and AMBR were proposed by Boicho Kokinov in the late 1980s and are still in active development. I joined the AMBR research group in the early 1990s as a graduate student in cognitive science at NBU.

I could not have asked for a better doctoral advisor than Boicho. He lavished his time and attention on me throughout my studies at NBU. We had countless conversations, discussing every single idea in this book. DUAL and AMBR are Boicho's creations and he should have been listed as first author. With his usual grace and modesty, however, he declined to have his name appear on the title page. This book is dedicated to him in appreciation of his mentorship, friendship, and support. Unfortunately, Boicho did not live long enough to see the published book but he did see the complete draft.

This book documents the status of the AMBR project as of 1998. I have resisted the temptation to revise the original dissertation text too much, although I did smooth out the roughest edges, particularly in Chapter 2 and Section 5.7. A new Afterword traces the development of DUAL and AMBR after 1998, whereas a lengthy new postscript to Chapter 2 provides pointers to the recent analogy-modeling literature at large. Appendix C is also new. The number of bibliographic references has more than tripled. Of course all faults in the text are entirely my responsibility. Corrections for errors discovered after the book goes to print will be posted on this site.

Although my research took a different turn after my graduation from NBU, I always kept a keen interest in analogy-making and followed the developments in this field. Cognitive science has advanced considerably in the intervening 15 years. Cognitive neuroscience in particular has made great strides and today we can incorporate much stronger neurological constraints into our models compared to 1998. I am currently developing an analogy-making model that is a neural network quite different from AMBR. But this is a story for another book.

This is a book about AMBR --- a treasure trove of Boicho's original ideas. AMBR definitely deserves careful study and is of potential interest to cognitive modelers and cognitive scientists more generally. Graduate students can use the text as an example of one possible way to write a Ph.D. thesis. Current members of the AMBR research group may be interested to learn about a precursor of their work.

I give heartfelt thanks to all who contributed to the work described here. To Boicho, first and foremost and always --- Thank you! I am forever in your debt and I will cherish the memory of you as long as I live. Rest in peace, my dear friend. To my professors and colleagues at the Central and Eastern European Center for Cognitive Science at NBU, who taught me so much and supported me in every way. As they are too many to enumerate here in full, I will single out just two names --- Encho Gerganov and Vassil Nikolov. To Bob French, whose book about the Tabletop model was never far from my desk while I was writing my thesis. Thank you, Bob, for the stimulating discussions and for your hospitality in the summer of 1996. To the members of my Dissertation Committee, who made time in their busy schedules during the First Analogy Conference in July 1998 to read and comment upon a long thesis --- Kenneth Forbus, Dedre Gentner, Keith Holyoak, John Hummel, Pentti Kanerva, Zdravko Markov, and Naum Yakimoff (chair). Thank you for your good will and encouragement. Your feedback and recommendations are now incorporated into the text. To Alexander Doumas, Keith Holyoak, John Hummel, Andrew Lovett, and Robert Morrison, for their critical comments on the Postscript to Chapter 2. To Georgi Petkov, who rose to the occasion and wrote the Afterword that Boicho meant to write but could not. To my mother and in loving memory of my father, who nurtured me with constant devotion and who were my first and most important teachers. Last but not least, very special thanks to my wonderful wife Petya and my adorable daughter Vicky.

Full text (pdf) Purchase on Amazon Kokinov & Petrov (2001)