From bozinovs@delusion.cs.umass.edu Mon Jan 1 03:54:52 1996 Received: from lucy.cs.wisc.edu by sea.cs.wisc.edu; Mon, 1 Jan 96 03:54:49 -0600; AA14770 Received: from TELNET-1.SRV.CS.CMU.EDU by lucy.cs.wisc.edu; Mon, 1 Jan 96 03:54:47 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id aa11449; 1 Jan 96 4:22:04 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id aa11447; 1 Jan 96 4:07:26 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa01332; 1 Jan 96 4:07:04 EST Received: from EDRC.CMU.EDU by B.GP.CS.CMU.EDU id aa26694; 31 Dec 95 18:05:06 EST Received: from delusion.cs.umass.edu by EDRC.CMU.EDU id aa18717; 31 Dec 95 18:04:55 EST Received: by delusion.cs.umass.edu (5.65/DEC-Ultrix/4.3) id AA25407; Sun, 31 Dec 1995 17:55:53 -0500 Date: Sun, 31 Dec 1995 17:55:53 -0500 From: bozinovs@delusion.cs.umass.edu Message-Id: <9512312255.AA25407@delusion.cs.umass.edu> To: Connectionists@cs.cmu.edu Subject: New Book Dear Connectionists, Happy New Year to everybody! At the end of the year I have a pleasure to announce a new book in the field. Advertisment: ********************************************************************* New Book! New Book! New Book! New Book! New Book! New Book! --------------------------------------------------------------------- CONSEQUENCE DRIVEN SYSTEMS CONSEQUENCE DRIVEN SYSTEMS CONSEQUENCE DRIVEN SYSTEMS by Stevo Bozinovski *201 pages *79 figures *27 algorithm descriptions *8 tables Among its special features, the book: --------------------------------------- ** provides a unified theory of response-sensitive teaching and learning ** as a result of that theory describes a generic architecture of a neuro-genetic agent capable of performing in 1) consequence sensitive teaching, 2) reinforcement learning, and 3) self-reinforcement learning paradigms ** describes the Crossbar Adaptive Array (CAA) architecture, an 1981 neural network developed within the Adaptive Networks Group, as an example of a neuro-genetic agent ** explains how the CAA architecture was the first neural network that solved a delayed reinforcement learning task, the Dungeons-and-Dragons task, in 1981 ** explains how the 1981 learning method (shown on the cover of the book) is actually the well known, 1989 rediscovered, Q-learning method ** introduces the Benefit-Cost CAA (B-C CAA), as extension of the 1981 Benefit-only CAA architecture ** introduces at-subgoal-go-back algorithm as modification of the 1981 at-goal-go-back CAA algorithm ** introduces a new type of neuron, denoted as Provoking Adaptive Unit, for dealing with tasks of Distributed Consequence Programming ** illustrates the usage of those neurons as routers in a routing-in-networks-with-faults task ** uses parallel programming technique in describing the algorithms throughout the book ----------------------------------------- Ordering information ISBN 9989-684-06-5, Gocmar Press, 1995 price: $15, paperback For further information contact the author: bozinovs@cs.umass.edu ********************************************************************** CONTENTS: 1. INTRODUCTION 1.1. The framework 1.2. Agents and architectures 1.3. Neural architectures 1.3.1. Greedy policy neural architectures 1.3.2. Recurrent architectures 1.3.3. Crossbar architectures 1.3.4. Subsumption architecture adaptive arrays 1.4. Problems. Emotional Graphs 1.5. Games. Emotional Petri Nets 1.6. Parallel programming 1.7. Bibliographical and other notes 2. CONSEQUENCE LEARNING AGENTS: A STRUCTURAL THEORY 2.1. The agent-environment interface 2.2. A taxonomy of learning paradigms 2.3. Classes of consequence learning agents 2.4. A generic consequence learning architecture 2.5. Learning rules and routines 2.6. Bibliographical and other notes 3. CONSEQUENCE DRIVEN TEACHING 3.1. Class T agents 3.2. Learners 3.3. Teachers 3.3.1. Toward a theory of teaching systems 3.3.2. Teaching strategies 3.4. Curriculums 3.4.1. Curriculum grammars and languages 3.4.2. Curriculum space approach 3.5. Pattern classification teaching as integer programming 3.6. Pattern classification teaching as dynamic programming 3.7. Bibliographical and other notes 4. EXTERNAL REINFORCEMENT LEARNING 4.1. Reinforcement learningh NG agents 4.2. Associative Search Network (ASN) 4.2.1. Basic ASN 4.2.2. Reionforcement predictive ASN 4.3. Actor-Critic architecture 4.4. Bibliographical and other notes 5. SELF-REINFORCEMENT LEARNING 5.1. Conceptual framework 5.2. Self-reinforcement learning and the NG agents 5.3. The Crossbar Adaptive Array architecture 5.4. How it works 5.4.1. Defining primary goals from the genetic environment 5.4.2. Secondary reinforcement mechanism 5.4.3. The CAA learning method 5.5. Example of a CAA architecture 5.6. Solving problems with a CAA architecture 5.6.1. Learning in emotional graphs: Maze running 5.6.2. Learning in loosely defined emotional graphs: Pole balancing 5.7. Another example of a CAA architecture 5.8. Using entropy in Markov Decision Processes 5.9. Issues on the genetic environment 5.9.1. CAA architecture as an optimization architecture 5.9.2. Complemetarity with the Genetic Algorithms 5.9.3. Self-reinforcement: Genetic environment approach 5.10. Bibliographical and other notes 6. CONSEQUENCE PROGRAMMING 6.1. Dynamic Programming and Markov Decision Problems 6.2. Introducing cost in the CAA architecture 6.3. Q-learning 6.4. A taxonomy of the CAA-method based learning algorithms 6.5. Producing optimal solution in a stochastic environment 6.6. Distributed Consequence Programming: A neural theory 6.6.1. Provoking units: Axon provoked neurons 6.6.2. An illustration: Routing in client-server networks with faults 6.7. Bibliographical and other notes 7. SUMMARY 8. REFERENCES 9. INDEX ********************************************************************* From koza@cs.stanford.edu Tue Jan 2 01:08:40 1996 Received: from lucy.cs.wisc.edu by sea.cs.wisc.edu; Tue, 2 Jan 96 01:08:38 -0600; AA22595 Received: from bilby.cs.uwa.oz.au by lucy.cs.wisc.edu; Tue, 2 Jan 96 01:08:36 -0600 Received: from (mafm@parma.cs.uwa.oz.au [130.95.1.7]) by cs.uwa.oz.au (8.6.8/8.5) with SMTP id NAA00844; Tue, 2 Jan 1996 13:20:55 +0800 Message-Id: <199601020520.NAA00844@cs.uwa.oz.au> From: "John R. Koza" To: Reinforce@cs.uwa.edu.au Subject: Univ. GA Course Information Book Date: Wed, 27 Dec 1995 19:49:46 -0800 (PST) NOW AVAILABLE!!! "THE GA 30" Information for Instructors and Prospective Instructors of University Courses on Genetic Algorithms TITLE: "University Courses on Genetic Algorithms 1995 Edition No. 1 - December, 1995 Compiled by John R. Koza, Computer Science Department, Stanford University This volume contains lightly-edited information about 30 different university courses on genetic algorithms that are offered by universities around the world. The information was contributed by the instructors of the various courses. This information was solicited by posting "requests for information" during 1995 on electronic mailing lists on genetic algorithms, genetic programming, and other topics related to evolutionary computation. It is hoped this collection will be useful to both instructors of existing courses on genetic algorithms and instructors considering starting up their own course on this subject. Copies of this volume (ISBN 0-18-195903P8) are available DIRECTLY from Stanford University Bookstore for $9.30 plus $6.00 shipping and handling (in the USA) by calling 415-329-1217 or 800-533-2670 or by writing Stanford Bookstore Stanford University Stanford, California 94305-3079 USA The E-Mail address of the bookstore for e-mail orders is mailorder@bookstore.stanford.edu. Be sure to mention the ISBN number, exact title, refer to "Custom Publishing" and "CSD 000" when ordering to avoid confusion with course readers, collections of student papers, and other materials associated with my courses at Stanford. John R. Koza Consulting Professor Computer Science Department Gates Building Stanford University Stanford, California 94305 USA PHONE: 415-941-0336 E-MAIL: Koza@Cs.Stanford.Edu WWW: http://www-cs-faculty.stanford.edu/~koza/ From koza@cs.stanford.edu Tue Jan 2 01:12:54 1996 Received: from lucy.cs.wisc.edu by sea.cs.wisc.edu; Tue, 2 Jan 96 01:12:52 -0600; AA22601 Received: from bilby.cs.uwa.oz.au by lucy.cs.wisc.edu; Tue, 2 Jan 96 01:12:47 -0600 Received: from (mafm@parma.cs.uwa.oz.au [130.95.1.7]) by cs.uwa.oz.au (8.6.8/8.5) with SMTP id NAA00968; Tue, 2 Jan 1996 13:24:19 +0800 Message-Id: <199601020524.NAA00968@cs.uwa.oz.au> From: "John R. Koza" To: Reinforce@cs.uwa.edu.au Subject: Book of 34 Student Papers Date: Wed, 27 Dec 1995 20:11:29 -0800 (PST) NOW AVAILABLE!!! A new collection of 34 student papers on GA and GP "Genetic Algorithms and Genetic Programming at Stanford 1995" Compiled by John R. Koza, Computer Science Department, Stanford University This volume (ISBN 0-18-195720-5) contains 34 papers written and submitted by students describing their term projects for the course "Genetic Algorithms and Genetic Programming" (Computer Science 426) at Stanford University offered during the fall quarter 1995 (both on campus and on SITN TV). The appendix to this volume contains material providing basic information about the course, including schedules, reading lists, project instructions, and the take-home final. In the take-home final examination in this course, each student "peer reviews" 4 papers written by other students in the class. Copies of the 1995 volume (ISBN 0-18-195720-5) are available DIRECTLY from Stanford University Bookstore for $14.96 plus $6.00 shipping and handling (in the USA) by calling 415-329-1217 or 800-533-2670 or by writing Stanford Bookstore Stanford University Stanford, California 94305-3079 USA The E-Mail address of the bookstore for e-mail orders is mailorder@bookstore.stanford.edu. Be sure to mention the ISBN number, exact title, refer to "Custom Publishing" and "CSD 000" when ordering to avoid confusion with course readers, collections of student papers, and other materials associated with my courses at Stanford. John R. Koza Consulting Professor Computer Science Department Gates Building Stanford University Stanford, California 94305 USA PHONE: 415-941-0336 E-MAIL: Koza@Cs.Stanford.Edu WWW: http://www-cs-faculty.stanford.edu/~koza/ ------------------------------------------------------------- TABLE OF CONTENTS Evolving Efficient Algorithms by Genetic Programming: A Case Study in Sorting by Eric T. Bauer Using Genetic Algorithms and Convolution to Find Optimal Strategies in Games without Perfect Information by Joey Beheler Genetic Fitting: Evolutionary Search of Optimal Approximations for Discrete Functions by Luca Benini Location Independent Pattern Recognition using Genetic Programming by Markus M. Breunig Valid English Word Classifier Using Genetic Programming by King Choi Chan Optimizing Local Area Networks Using Genetic Algorithms by Andy Choi Predator-Prey Interactions in a Simulated World by Adam Clark Evolution of General Algorithmic Solutions for Simple Sliding Tile Puzzles by Thomas Dillon Evolving Effective Solutions in Effective Amounts of Time by David Engel The Application of Genetic Programming to Cooperative Movement Planning and Execution by John Hart Genetic Programming of Near Minimum Time Spacecraft Attitude Maneuvers by Brian Howley A Genetic Algorithm for a Stochastic Network Planning Problem by David Joffe An Attempt to Evolve Cooperation Among Separately Evolved Structure in Genetic Programming by Bryan H. Johnson Error Driven Parallelization of a Genetic Program by Sesha Kalyur Behavior Learning and Individual Cooperation in Autonomous Agents as a Result of Interaction Dynamics with the Environment by Sejal Kamani The Genetically Determined Dream Team by Mark Kanok A Variable Complexity Genetic Algorithm for Job Allocation by Sanjay Kapoor Using Genetic Algorithm and Decision Trees to produce a Hybrid Classification System by D'ondria L. Kennard Development of Navigational Controllers for Vehicles in Highway Traffic Situations via Genetic Programming by Lisa A. Laane Camera Placement for Optimal Visibility by Vui Chiap Lam The Genetic Algorithm applied to Gate Sizing by Jeremy R. Levitt An Evolutionary Approach to CPU Fault Isolation by Keith Mac Donald Emergent Behavior in Traffic Light Controllers using Genetic Programming by Ari W. Mozes The Hannibal Project by Carl Orthlieb On the Use of Genetic Programming in Elevator Control Design by Dan Pietrasik Evolution of Communication and Division of Labor via Genetic Programming by Hanno Sander Genetic Algorithms Applied to Machine Language by Christian R. Shelton An Empirical Comparison of 3 Population-Based Search Algorithms for the Traveling Salesman Problem by Sanjeev Singh Discovering Patterns in Two-Dimensional Cellular Automata by Caz Taylor Are Your Ready for Some Football? Genetically Produced Ratings for College Football Teams by Howard Thompson Recognition and Reconstruction of Visibility Graphs Using a Genetic Algorithm by Marshall S. Veach Genetic Evolution of Behavior-Oriented Robots by Thomas Willeke Playing Tetris Using Genetic Programming by Michael Yurovitsky Genetic Algorithms in the Solution of Assembly Line Balancing Problems by Greg Zaric Appendix containing materials about the course ------------------------------------------------------------- ALSO AVAILABLE Contact the bookstore directly for current prices on these past items: --- Genetic Algorithms at Stanford 1994 (ISBN 0-18- 187263-3) P 20 papers from the fall quarter 1994. --- Artificial Life at Stanford 1994 (ISBN 0-18-182105-2) P 22 papers from the spring quarter 1994. --- Artificial Life at Stanford 1993 (ISBN 0-18-171957-6) --- Genetic Algorithms at Stanford 1993 (ISBN 0-18- 1738252). --- A course reader entitled Course Reader for Computer Science 426 (Genetic Algorithms) for Fall Quarter 1995 (ISBN 0-18-192183-9) contains 10 selected papers from the current genetic algorithms and genetic programming literature to supplement the two textbooks used in the CS 426 course. From mpolycar@ece.uc.edu Tue Jan 2 15:00:55 1996 Received: from lucy.cs.wisc.edu by sea.cs.wisc.edu; Tue, 2 Jan 96 15:00:53 -0600; AA28184 Received: from TELNET-1.SRV.CS.CMU.EDU by lucy.cs.wisc.edu; Tue, 2 Jan 96 15:00:50 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id aa16479; 2 Jan 96 14:21:19 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id aa16465; 2 Jan 96 13:58:51 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa01871; 2 Jan 96 13:58:44 EST Received: from RI.CMU.EDU by B.GP.CS.CMU.EDU id aa11437; 2 Jan 96 10:20:51 EST Received: from babbage.ece.uc.edu by RI.CMU.EDU id aa28913; 2 Jan 96 10:20:25 EST Received: from zoe.ece.uc.edu (mpolycar@zoe.ece.uc.edu [129.137.8.203]) by babbage.ece.uc.edu (8.7.2/8.7.3) with ESMTP id KAA17559; Tue, 2 Jan 1996 10:19:37 -0500 (EST) From: Marios Polycarpou Received: (mpolycar@localhost) by zoe.ece.uc.edu (8.6.9/8.6.4) id KAA19208; Tue, 2 Jan 1996 10:19:35 -0500 Message-Id: <199601021519.KAA19208@zoe.ece.uc.edu> Subject: ISIC'96: Final Call for Papers To: Connectionists@cs.cmu.edu Date: Tue, 2 Jan 1996 10:19:35 -0500 (EST) X-Mailer: ELM [version 2.4 PL23] Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit FINAL CALL FOR PAPERS 11th IEEE International Symposium on Intelligent Control (ISIC'96) Sponsored by the IEEE Control Systems Society and held in conjunction with The 1996 IEEE International Conference on Control Applications (CCA) and The IEEE Symposium on Computer-Aided Control System Design (CACSD) September 15-18, 1996 The Ritz-Carlton Hotel, Dearborn, Michigan, USA ISIC General Chair: Kevin M. Passino, The Ohio State University ISIC Program Chair: Jay A. Farrell, University of California, Riverside ISIC Publicity Chair: Marios Polycarpou, University of Cincinnati Intelligent control, the discipline where control algorithms are developed by emulating certain characteristics of intelligent biological systems, is being fueled by recent advancements in computing technology and is emerging as a technology that may open avenues for significant technological advances. For instance, fuzzy controllers which provide for a simplistic emulation of human deduction have been heuristically constructed to perform difficult nonlinear control tasks. Knowledge-based controllers developed using expert systems or planning systems have been used for hierarchical and supervisory control. Learning controllers, which provide for a simplistic emulation of human induction, have been used for the adaptive control of uncertain nonlinear systems. Neural networks have been used to emulate human memorization and learning characteristics to achieve high performance adaptive control for nonlinear systems. Genetic algorithms that use the principles of biological evolution and "survival of the fittest" have been used for computer-aided-design of control systems and to automate the tuning of controllers by evolving in real-time populations of highly fit controllers. Topics in the field of intelligent control are gradually evolving, and expanding on and merging with those of conventional control. For instance, recent work has focused on comparative cost-benefit analyses of conventional and intelligent control techniques using simulation and implementations. In addition, there has been recent activity focused on modeling and nonlinear analysis of intelligent control systems, particularly work focusing on stability analysis. Moreover, there has been a recent focus on the development of intelligent and conventional control systems that can achieve enhanced autonomous operation. Such intelligent autonomous controllers try to integrate conventional and intelligent control approaches to achieve levels of performance, reliability, and autonomous operation previously only seen in systems operated by humans. Papers are being solicited for presentation at ISIC and for publication in the Symposium Proceedings on topics such as: - Architectures for intelligent control - Hierarchical intelligent control - Distributed intelligent systems - Modeling intelligent systems - Mathematical analysis of intelligent systems - Knowledge-based systems - Fuzzy systems / fuzzy control - Neural networks / neural control - Machine learning - Genetic algorithms - Applications / Implementations: - Automotive / vehicular systems - Robotics / Manufacturing - Process control - Aircraft / spacecraft This year the ISIC is being held in conjunction with the 1996 IEEE International Conference on Control Applications and the IEEE Symposium on Computer-Aided Control System Design. Effectively this is one large conference at the beautiful Ritz-Carlton hotel. The programs will be held in parallel so that sessions from each conference can be attended by all. There will be one registration fee and each registrant will receive a complete set of proceedings. For more information, and information on how to submit a paper to the conference see the back of this sheet. ++++++++++ Submissions: ++++++++++ Papers: Five copies of the paper (including an abstract) should be sent by Jan. 22, 1996 to: Jay A. Farrell, ISIC'96 College of Engineering ph: (909) 787-2159 University of California, Riverside fax: (909) 787-3188 Riverside, CA 92521 Jay_Farrell@qmail.ucr.edu Clearly indicate who will serve as the corresponding author and include a telephone number, fax number, email address, and full mailing address. Authors will be notified of acceptance by May 1996. Accepted papers, in final camera ready form (maximum of 6 pages in the proceedings), will be due in June 1996. Invited Sessions: Proposals for invited sessions are being solicited and are due Jan. 22, 1996. The session organizers should contact the Program Chair by Jan. 1, 1996 to discuss their ideas and obtain information on the required invited session proposal format. Workshops and Tutorials: Proposals for pre-symposium workshops should be submitted by Jan. 22, 1996 to: Kevin M. Passino, ISIC'96 Dept. Electrical Engineering ph: (614) 292-5716 The Ohio State University fax: (614) 292-7596 2015 Neil Ave. passino@osu.edu Columbus, OH 43210-1272 Please contact K.M. Passino by Jan. 1, 1996 to discuss the content and required format for the workshop or tutorial proposal. ++++++++++++++++++++++++ Symposium Program Committee: ++++++++++++++++++++++++ James Albus, National Institute of Standards and Technology Karl Astrom, Lund Institute of Technology Matt Barth, University of California, Riverside Michael Branicky, Massachusetts Institute of Technology Edwin Chong, Purdue University Sebastian Engell, University of Dortmund Toshio Fukuda, Nagoya University Zhiqiang Gao, Cleveland State University Dimitry Gorinevsky, Measurex Devron Inc. Ken Hunt, Daimler-Benz AG Tag Gon Kim, KAIST Mieczyslaw Kokar, Northeastern University Ken Loparo, Case Western Reserve University Kwang Lee, The Pennsylvania State University Michael Lemmon, University of Notre Dame Frank Lewis, University of Texas at Arlington Ping Liang, University of California, Riverside Derong Liu, General Motors R&D Center Kumpati Narendra, Yale University Anil Nerode, Cornell University Marios Polycarpou, University of Cincinnati S. Joe Qin, Fisher-Rosemount Systems, Inc. Tariq Samad, Honeywell Technology Center George Saridis, Rensselaer Polytechnic Institute Jennie Si, Arizona State University Mark Spong, University of Illinois at Urbana-Champaign Jeffrey Spooner, Sandia National Laboratories Harry Stephanou, Rensselaer Polytechnic Institute Kimon Valavanis, University of Southwestern Louisiana Li-Xin Wang, Hong Kong University of Science and Tech. Gary Yen, USAF Phillips Laboratory ************************************************************************** * Prof. Marios M. Polycarpou | TEL: (513) 556-4763 * * University of Cincinnati | FAX: (513) 556-7326 * * Dept. Electrical & Computer Engineering | * * Cincinnati, Ohio 45221-0030 | Email: polycarpou@uc.edu * ************************************************************************** From wilson@smith.rowland.org Wed Jan 3 19:47:29 1996 Received: from lucy.cs.wisc.edu by sea.cs.wisc.edu; Wed, 3 Jan 96 19:47:22 -0600; AA04037 Received: from bilby.cs.uwa.oz.au by lucy.cs.wisc.edu; Wed, 3 Jan 96 19:47:19 -0600 Received: from (mafm@parma.cs.uwa.oz.au [130.95.1.7]) by cs.uwa.oz.au (8.6.8/8.5) with SMTP id IAA14975; Thu, 4 Jan 1996 08:21:46 +0800 Message-Id: <199601040021.IAA14975@cs.uwa.oz.au> From: Stewart Wilson To: reinforce@cs.uwa.edu.au Subject: Sutton article on NetQ Date: Tue, 02 Jan 1996 16:11:30 -0500 Rich Sutton's article "Generalization in Reinforcement Learning: Successful Examples Using Sparse Coarse Coding" (NIPS'95) has been added to NetQ (http://netq.rowland.org). NetQ provides a way to question an author about his paper (anonymously), and to read answers to other people's questions. Rich's article is the first strictly RL article on NetQ, though at least one other is RL-related. --Stewart Wilson From hicks@cs.titech.ac.jp Thu Jan 4 10:57:50 1996 Received: from lucy.cs.wisc.edu by sea.cs.wisc.edu; Thu, 4 Jan 96 10:57:47 -0600; AA00941 Received: from TELNET-1.SRV.CS.CMU.EDU by lucy.cs.wisc.edu; Wed, 3 Jan 96 23:13:39 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id aa20015; 3 Jan 96 23:06:45 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id aa20013; 3 Jan 96 22:46:31 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa05365; 3 Jan 96 22:45:54 EST Received: from CS.CMU.EDU by B.GP.CS.CMU.EDU id aa28812; 3 Jan 96 19:59:48 EST Received: from euclid.cs.titech.ac.jp by CS.CMU.EDU id aa26082; 3 Jan 96 19:59:01 EST Received: by euclid.cs.titech.ac.jp (8.6.11+2.4W/3.2W-titmx-2.0g); Thu, 4 Jan 1996 09:57:57 +0900 Date: Thu, 4 Jan 1996 09:57:57 +0900 From: hicks@cs.titech.ac.jp Message-Id: <199601040057.JAA27152@euclid.cs.titech.ac.jp> To: connectionists@cs.cmu.edu Subject: Query: Infinite priors I have the following question concerning the existence of certain priors. Can we say that a uniform prior over an infinite domain exists? For example, the uniform prior over all natural numbers. I wonder since it cannot be expressed in the form p(n) = (f(n)/\sum_n f(n)), where f(n) is a well defined function over the natural numbers. In general, if f(n) is not a summable series, then can the probability function p(n), whose elements have the ratios of the elements f(n), i.e., p(n)/p(m) = f(n)/f(m), be said to exist? I ask because a true Bayesian approach to some problems may require the prior to be defined. If there is no prior, then we can't say we are taking a Bayesian approach. If an infinite uniform prior does not exist, then we cannot take the approach that "no prior knowledge" = "infinite uniform prior". I.e., it would imply that any Bayesian approach involving the prior MUST begin with some assumptions about the prior (i.e., it must be formed from a summable/integrable function). References or opinions would be welcome. Craig Hicks. hicks@cs.titech.ac.jp From finnoff@predict.com Thu Jan 4 16:46:15 1996 Received: from lucy.cs.wisc.edu by sea.cs.wisc.edu; Thu, 4 Jan 96 16:46:12 -0600; AA07861 Received: from TELNET-1.SRV.CS.CMU.EDU by lucy.cs.wisc.edu; Thu, 4 Jan 96 16:46:10 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id aa20795; 4 Jan 96 15:05:09 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id aa20783; 4 Jan 96 14:47:33 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa06118; 4 Jan 96 14:47:08 EST Received: from CS.CMU.EDU by B.GP.CS.CMU.EDU id aa07343; 4 Jan 96 13:36:45 EST Received: from GRANDE.NM.ORG by CS.CMU.EDU id aa02407; 4 Jan 96 13:36:19 EST Received: from predict.com by grande.nm.org with SMTP; Thu, 4 Jan 1996 11:36:23 -0700 (MST) Received: by predict.com (5.x/BK-1.2) id AA27205; Thu, 4 Jan 1996 11:34:32 -0700 >Return-Path: Received: from mule.predict.com by predict.com (4.1/BK-1.2) id AA08352; Thu, 4 Jan 96 11:35:37 MST Date: Thu, 4 Jan 96 11:35:37 MST From: William Finnoff Message-Id: <9601041835.AA08352@predict.com> To: connectionists@cs.cmu.edu Subject: Re: Query: Infinite priors Content-Type: text Craig Hicks writes: > I have the following question concerning the existence of certain priors. > Can we say that a uniform prior over an infinite domain exists? As Craig noted, there is no "uniform" probability measure on a countably infinite space which has the power set as its sigma algebra. What one is looking for is something that is a generalization of the uniform measure on a space of finite cardinality. What properties of this measure does one want to have on an extension to the infinite case? That really depends on use of the extended "uniform" measure, but several possibilities are: 1) Each point has the same probability of occurring. 2) The measure is invariant to any one to one mapping (permutation) of the points in the space. 3) The measure has maximum entropy. On an infinite space with uncountably infinite sigma algbra there does not exist a unique probability measure that has any one of these properties. In the first case, uniqueness fails when nonatomic measures are possible (for example for uncountably infinite spaces). In the second case, only a nonfinite measure can produce the desired results for a countably infinite space with the power set as sigma algebra. In the third case, it is not even possible to define the property without some "reference measure" on the space with respect to which entropy can be defined. Then one must ask: "What is an appropriate reference measure?" There is a long history and extensive literature in Bayesian statistics dealing with the search for "non-informative" prior distributions that in some way appropriately approximate desired properties of the uniform distribution on a finite space. An excellent discussion of the history of this problem and an extensive list of references can be found in section 5.6.2 of: Bernardo, J.M. and A.F.M. Smith, "Bayesian Theory", John Wiley and Sons, New York, 1994. Hope this helps you out. William %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% William Finnoff Prediction Co. 320 Aztec St., Suite B Santa Fe, NM, 87501, USA Tel.: (505)-984-3123 Fax: (505)-983-0571 e-mail: finnoff@predict.com From om@research.nj.nec.com Thu Jan 4 16:46:16 1996 Received: from lucy.cs.wisc.edu by sea.cs.wisc.edu; Thu, 4 Jan 96 16:46:15 -0600; AA07867 Received: from TELNET-1.SRV.CS.CMU.EDU by lucy.cs.wisc.edu; Thu, 4 Jan 96 16:46:12 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id ab20795; 4 Jan 96 15:05:51 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id ab20783; 4 Jan 96 14:47:34 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa06122; 4 Jan 96 14:47:11 EST Received: from EDRC.CMU.EDU by B.GP.CS.CMU.EDU id aa07715; 4 Jan 96 14:00:32 EST Received: from zingo.nj.nec.com by EDRC.CMU.EDU id aa09054; 4 Jan 96 13:59:38 EST Received: by zingo (940816.SGI.8.6.9/YDL1.4-910307.16) id NAA05525(zingo); Thu, 4 Jan 1996 13:56:33 -0500 Received: by iris64 (931110.SGI/cliff's joyful mailer #2) id AA01276(iris64); Thu, 4 Jan 96 13:56:31 -0500 Date: Thu, 4 Jan 96 13:56:31 -0500 From: "Stephen M. Omohundro" Message-Id: <9601041856.AA01276@iris64> To: hicks@cs.titech.ac.jp Cc: connectionists@cs.cmu.edu In-Reply-To: <199601040057.JAA27152@euclid.cs.titech.ac.jp> (hicks@cs.titech.ac.jp) Subject: Re: Query: Infinite priors > Date: Thu, 4 Jan 1996 09:57:57 +0900 > From: hicks@cs.titech.ac.jp > > > I have the following question concerning the existence of certain priors. > > Can we say that a uniform prior over an infinite domain exists? For example, > the uniform prior over all natural numbers. I wonder since it cannot be > expressed in the form p(n) = (f(n)/\sum_n f(n)), where f(n) is a well defined > function over the natural numbers. In general, if f(n) is not a summable > series, then can the probability function p(n), whose elements have the ratios > of the elements f(n), i.e., p(n)/p(m) = f(n)/f(m), be said to exist? > > I ask because a true Bayesian approach to some problems may require the prior > to be defined. If there is no prior, then we can't say we are taking a > Bayesian approach. If an infinite uniform prior does not exist, then we > cannot take the approach that "no prior knowledge" = "infinite uniform prior". > I.e., it would imply that any Bayesian approach involving the prior MUST begin > with some assumptions about the prior (i.e., it must be formed from a > summable/integrable function). > > References or opinions would be welcome. Craig Hicks. hicks@cs.titech.ac.jp > These are generally called "improper priors". You can still do much of the Bayesian paradigm using them because in many situations the likelihood function is such that the posterior (which is proportional to the prior times the likelihood) is normalizable even if the prior isn't. Formally you can treat them using a limiting sequence of proper priors. Most books on Bayesian analysis have some discussion of this topic. --Steve -- Stephen M. Omohundro http://www.neci.nj.nec.com/homepages/om NEC Research Institute, Inc. om@research.nj.nec.com 4 Independence Way Phone: 609-951-2719 Princeton, New Jersey 08540 Fax: 609-951-2488 From steven.young@psy.ox.ac.uk Thu Jan 4 23:19:11 1996 Received: from lucy.cs.wisc.edu by sea.cs.wisc.edu; Thu, 4 Jan 96 23:19:07 -0600; AA11046 Received: from TELNET-1.SRV.CS.CMU.EDU by lucy.cs.wisc.edu; Thu, 4 Jan 96 23:19:06 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id aa20781; 4 Jan 96 14:57:26 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id aa20779; 4 Jan 96 14:46:42 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa06108; 4 Jan 96 14:45:57 EST Received: from RI.CMU.EDU by B.GP.CS.CMU.EDU id aa05433; 4 Jan 96 10:48:34 EST Received: from [163.1.2.9] by RI.CMU.EDU id aa10531; 4 Jan 96 10:47:57 EST Received: from cogsci1.psych.ox.ac.uk by oxmail3 with SMTP (PP); Thu, 4 Jan 1996 15:47:14 +0000 Received: (from young@localhost) by cogsci1.psych.ox.ac.uk (8.6.11/8.6.12) id PAA13854; Thu, 4 Jan 1996 15:47:00 GMT From: Steven Young Message-Id: <199601041547.PAA13854@cogsci1.psych.ox.ac.uk> Subject: LAST CALL for participation for the Oxford Summer School on Connectionist Modelling To: cogling@ucsd.edu, connectionists@cs.cmu.edu, dev-europe@durham.ac.uk, info-childes@poppy.psy.cmu.edu, neuron@cattell.psych.upenn.edu, psyc@pucc.princeton.edu, psyling@psy.gla.ac.uk Date: Thu, 4 Jan 1996 15:46:59 +0000 (GMT) Cc: Steven.Young@psy.ox.ac.uk, Kim.Plunkett@psy.ox.ac.uk, Susan.King@psy.ox.ac.uk X-Mailer: ELM [version 2.4 PL24] Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Content-Length: 2528 This is the LAST CALL for participation for the 1996 Oxford Summer School on Connectionist Modelling follows. Please pass on this information to people you know who would be interested. -------- OXFORD SUMMER SCHOOL ON CONNECTIONIST MODELLING Department of Experimental Psychology University of Oxford 21 July - 2nd August 1996 Applications are invited for participation in a 2-week residential Summer School on techniques in connectionist modelling. The course is aimed primarily at researchers who wish to exploit neural network models in their teaching and/or research and it will provide a general introduction to connectionist modelling through lectures and exercises on Power PCs. The course is interdisciplinary in content though many of the illustrative examples are taken from cognitive and developmental psychology, and cognitive neuroscience. The instructors with primary responsibility for teaching the course are Kim Plunkett and Edmund Rolls. No prior knowledge of computational modelling will be required though simple word processing skills will be assumed. Participants will be encouraged to start work on their own modelling projects during the Summer School. The cost of participation in the Summer School is #750 to include accommodation (bed and breakfast at St. John's College) and registration. Participants will be expected to cover their own travel and meal costs. A small number of graduate student scholarships providing partial funding may be available. Applicants should indicate whether they wish to be considered for a graduate student scholarship but are advised to seek their own funding as well, since in previous years the number of graduate student applications has far exceeded the number of scholarships available. There is a Summer School World Wide Web page describing the contents of the 1995 Summer School available on: http://cogsci1.psych.ox.ac.uk/summer-school/ Further information about contents of the course can be obtained from Steven.Young@psy.ox.ac.uk If you are interested in participating in the Summer School, please contact: Mrs Sue King Department of Experimental Psychology University of Oxford South Parks Road Oxford OX1 3UD Tel: (01865) 271353 Email: susan.king@psy.oxford.ac.uk Please send a brief description of your background with an explanation of why you would like to attend the Summer School (one page maximum) no later than 31st January 1996. Regards, Steven Young. From radford@cs.toronto.edu Thu Jan 4 23:19:13 1996 Received: from lucy.cs.wisc.edu by sea.cs.wisc.edu; Thu, 4 Jan 96 23:19:09 -0600; AA11048 Received: from TELNET-1.SRV.CS.CMU.EDU by lucy.cs.wisc.edu; Thu, 4 Jan 96 23:19:08 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id ac20795; 4 Jan 96 15:06:17 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id ac20783; 4 Jan 96 14:47:35 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa06126; 4 Jan 96 14:47:15 EST Received: from CS.CMU.EDU by B.GP.CS.CMU.EDU id aa07945; 4 Jan 96 14:16:30 EST Received: from yonge.cs.toronto.edu by CS.CMU.EDU id aa02765; 4 Jan 96 14:16:23 EST Received: from neuron.ai.toronto.edu ([128.100.3.14]) by yonge.cs.toronto.edu with SMTP id <86532>; Thu, 4 Jan 1996 14:16:01 -0500 Received: by neuron.ai.toronto.edu id <965>; Thu, 4 Jan 1996 14:15:37 -0500 From: Radford Neal To: connectionists@cs.cmu.edu Subject: Re: Query on "infinite" priors Cc: radford@cs.toronto.edu Message-Id: <96Jan4.141537edt.965@neuron.ai.toronto.edu> Date: Thu, 4 Jan 1996 14:15:27 -0500 Craig Hicks. hicks@cs.titech.ac.jp writes: > Can we say that a uniform prior over an infinite domain exists? For > example, the uniform prior over all natural numbers... > > I ask because a true Bayesian approach to some problems may require > the prior to be defined. If there is no prior, then we can't say we > are taking a Bayesian approach. If an infinite uniform prior does not > exist, then we cannot take the approach that "no prior knowledge" = > "infinite uniform prior". I.e., it would imply that any Bayesian > approach involving the prior MUST begin with some assumptions about > the prior (i.e., it must be formed from a summable/integrable function). This is a long-standing issue in Bayesian inference. These "infinite" priors are usually called "improper" priors, while those that can be normalized are called "proper" priors. Some Bayesians like to use improper priors, as long as the posterior turns out to be proper (which is often, but not always, the case). Other Bayesians eschew improper priors, because strange things can sometimes occur when you use them. One strangeness is that a Bayesian procedure based on an improper prior can be "inadmissible" - ie, be uniformly worse than some other procedure with respect to expected performance, for any state of the world. A famous example (Stein's paradox) is estimation of the mean of a vector of three or more independent components having Gaussian distributions, with the aim of minimizing the expected squared error. The Bayesian estimate with an improper uniform prior is just the sample mean, which turns out to be inadmissible. In contrast Bayesian procedures based on proper priors that are nowhere zero are always admissible. There should be lots of stuff on this in Bayesian textbooks, such as Smith and Bernardo's recent book on "Bayesian Theory" (though I don't have a copy handy to verify just what they say). ---------------------------------------------------------------------------- Radford M. Neal radford@cs.toronto.edu Dept. of Statistics and Dept. of Computer Science radford@utstat.toronto.edu University of Toronto http://www.cs.toronto.edu/~radford ---------------------------------------------------------------------------- From plunkett@crl.ucsd.edu Thu Jan 4 23:19:13 1996 Received: from lucy.cs.wisc.edu by sea.cs.wisc.edu; Thu, 4 Jan 96 23:19:12 -0600; AA11056 Received: from TELNET-1.SRV.CS.CMU.EDU by lucy.cs.wisc.edu; Thu, 4 Jan 96 23:19:09 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id aa20998; 4 Jan 96 17:39:16 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id aa20996; 4 Jan 96 17:30:44 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa06236; 4 Jan 96 17:30:16 EST Received: from CS.CMU.EDU by B.GP.CS.CMU.EDU id ab06047; 4 Jan 96 11:41:52 EST Received: from crl.ucsd.edu by CS.CMU.EDU id aa01205; 4 Jan 96 11:41:16 EST Received: by crl.ucsd.edu; id AA17397 sendmail 4.1/UCSD-2.1-sun Thu, 4 Jan 96 08:41:46 PST for connectionists@cs.cmu.edu Date: Thu, 4 Jan 96 08:41:46 PST From: Kim Plunkett Message-Id: <9601041641.AA17397@crl.ucsd.edu> To: connectionists@cs.cmu.edu University Lectureship University of Oxford Department of Experimental Psychology Applications are invited from human experimental/cognitive psychologists (including connectionist modellers) with a proven record of research and training. The post is tenable from 1 october 1996 or as soon as possible thereafter. The stipend will be according to age on the scale stlg15,154-stlg28,215 per annum. The successful candidate may be offered an Official Fellowship at New College, for which additional renumeration would be available. Further particulars, containing details of the duties and the full range of emoluments and allowances attaching to both the University and College appointments, may be obtained from Professor S.D. Iversen, Department of Experimental Psychology, South Parks Road, Oxford OX1 3UD, U.K. (telephone +44 (0) 1865 271356; fax +44 (0) 1865 271354) to applications (eight copies, two only from overseas candidates), containing a curriculum vitae, a summary of research, a list of principal publications and the names of three referees, should be sent by 31 January 1996. Candidates will be notified if they are required for interview. The University exists to promote excellence in education and research, and is an equal opportunities employer. From ali@almaden.ibm.com Fri Jan 5 22:40:11 1996 Received: from lucy.cs.wisc.edu by sea.cs.wisc.edu; Fri, 5 Jan 96 22:40:08 -0600; AA26385 Received: from TELNET-1.SRV.CS.CMU.EDU by lucy.cs.wisc.edu; Fri, 5 Jan 96 22:40:07 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id aa23046; 5 Jan 96 22:13:53 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id aa23044; 5 Jan 96 22:05:46 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa07601; 5 Jan 96 22:05:24 EST Received: from CS.CMU.EDU by B.GP.CS.CMU.EDU id aa29069; 5 Jan 96 20:43:51 EST Received: from [206.65.71.130] by CS.CMU.EDU id aa13864; 5 Jan 96 20:43:17 EST Received: from ALMADEN by almaden.ibm.com (IBM VM SMTP V2R2) with BSMTP id 0045; Fri, 05 Jan 96 17:41:51 PST Received: by ALMADEN (XAGENTA 4.0) id 2315; Fri, 5 Jan 1996 17:41:50 -0800 Received: by brasil.almaden.ibm.com (AIX 4.1/UCB 5.64/4.03) id AA21382; Fri, 5 Jan 1996 17:43:02 -0800 Date: Fri, 5 Jan 1996 17:43:02 -0800 From: ali@almaden.ibm.com Mmdf-Warning: Parse error in original version of preceding line at DST.BOLTZ.CS.CMU.EDU Message-Id: <9601060143.AA21382@brasil.almaden.ibm.com> To: connectionists@cs.cmu.edu Subject: Dissertation announcement The following dissertation is available via anonymous FTP and through http://www.ics.uci.edu/~ali (either as a whole or by chapters). Title: "Learning Probabilistic Relational Concept Descriptions" By Kamal Ali Key words: Learning probabilistic concepts, multiple models, multiple classifiers, combining classifiers, evidence combination, relational learning, First-order learning, Noise-tolerant learning, Learning of small disjuncts, Inductive Logic Programming. A B S T R A C T This dissertation presents results in the area of multiple models (multiple classifiers), learning probabilistic relational (first order) rules from noisy, "real-world" data and reducing the small disjuncts problem - the problem whereby learned rules that cover few training examples have high error rates on test data. Several results are presented in the arena of multiple models. The multiple models approach in relevant to the problem of making accurate classifications in ``real-world'' domains since it facilitates evidence combination which is needed to accurately learn on such domains. It is also useful when learning from small training data samples in which many models appear to be equally "good" w.r.t. the given evaluation metric. Such models often have quite varying error rates on test data so in such situations, the single model method has problems. Increasing search only partly addresses this problem whereas the multiple models approach has the potential to be much more useful. The most important result of the multiple models research is that the *amount* of error reduction afforded by the multiple models approach is linearly correlated with the degree to which the individual models make errors in an uncorrelated manner. This work is the first to model the degree of error reduction due to the use of multiple models. It is also shown that it is possible to learn models that make less correlated errors in domains in which there are many ties in the search evaluation metric during learning. The third major result of the research on multiple models is the realization that models should be learned that make errors in a negatively-correlated manner rather than those that make errors in an uncorrelated (statistically independent) manner. The thesis also presents results on learning probabilistic first-order rules from relational data. It is shown that learning a class description for each class in the data - the one-per-class approach - and attaching probabilistic estimates to the learned rules allows accurate classifications to be made on real-world data sets. The thesis presents the system HYDRA which implements this approach. It is shown that the resulting classifications are often more accurate than those made by three existing methods for learning from noisy, relational data. Furthermore, the learned rules are relational and so are more expressive than the attribute-value rules learned by most induction systems. Finally, results are presented on the small-disjuncts problem in which rules that apply to rare subclasses have high error rates The thesis presents the first approach that is simultaneously successful at reducing the error rates of small disjucnts while also reducing the overall error rate by a statistically significant margin. The previous approach which aimed to reduce small disjunct error rates only did so at the expense of increasing the error rates of large disjuncts. It is shown that the one-per-class approach reduces error rates for such rare rules while not sacrificing the error rates of the other rules. The dissertation is approximately 180 pages long (single spaced) (~590K). ftp ftp.ics.uci.edu logname: anonymous password: your email address cd /pub/ali binary get thesis.ps.Z quit ============================================================================ I am now with the IBM Data Mining group at Almaden (San Jose) - we are looking for good people for data analysis (data mining) and consulting so please feel free to call me at (408) 365 8736. My address is: Kamal Ali, Room D3-250 IBM Almaden Research Center 650 Harry Rd San Jose, CA 95120 ============================================================================== Kamal Mahmood Ali, Ph.D. Phone: 408 927 1354 Consultant and data mining analyst, Fax: 408 927 3025 Data Mining Solutions, Office: ARC D3-250 IBM http://www.almaden.ibm.com/stss/ ============================================================================== From rao@cs.rochester.edu Sat Jan 6 20:39:31 1996 Received: from lucy.cs.wisc.edu by sea.cs.wisc.edu; Sat, 6 Jan 96 20:39:29 -0600; AA03329 Received: from TELNET-1.SRV.CS.CMU.EDU by lucy.cs.wisc.edu; Sat, 6 Jan 96 20:39:27 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id aa24125; 6 Jan 96 20:28:07 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id aa24123; 6 Jan 96 20:11:06 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa08351; 6 Jan 96 20:10:57 EST Received: from EDRC.CMU.EDU by B.GP.CS.CMU.EDU id aa09458; 6 Jan 96 18:26:13 EST Received: from cayuga.cs.rochester.edu by EDRC.CMU.EDU id aa17307; 6 Jan 96 18:26:07 EST Received: from vulture.cs.rochester.edu (vulture.cs.rochester.edu [192.5.53.47]) by cayuga.cs.rochester.edu (8.6.9/H) with ESMTP id SAA14073; Sat, 6 Jan 1996 18:25:58 -0500 Received: (from rao@localhost) by vulture.cs.rochester.edu (8.6.9/K) id SAA14449; Sat, 6 Jan 1996 18:25:56 -0500 Date: Sat, 6 Jan 1996 18:25:56 -0500 From: rao@cs.rochester.edu Message-Id: <199601062325.SAA14449@vulture.cs.rochester.edu> To: Connectionists@cs.cmu.edu, neuron@CATTELL20.psych.upenn.edu Subject: Paper Available: Eye Movements in Visual Search Modeling Saccadic Targeting in Visual Search Rajesh P.N. Rao, Gregory J. Zelinsky, Mary M. Hayhoe and Dana H. Ballard Department of Computer Science University of Rochester Rochester, NY 14627, USA To appear in [Advances in Neural Information Processing Systems 8 (NIPS*95), D. Touretzky, M. Mozer and M. Hasselmo (Eds.), MIT Press, 1996] Abstract Visual cognition depends critically on the ability to make rapid eye movements known as saccades that orient the fovea over targets of interest in a visual scene. Saccades are known to be ballistic: the pattern of muscle activation for foveating a prespecified target location is computed prior to the movement and visual feedback is precluded. Despite these distinctive properties, there has been no general model of the saccadic targeting strategy employed by the human visual system during visual search in natural scenes. This paper proposes a model for saccadic targeting that uses iconic scene representations derived from oriented spatial filters at multiple scales. Visual search proceeds in a coarse-to-fine fashion with the largest scale filter responses being compared first. The model was empirically tested by comparing its performance with actual eye movement data from human subjects in a natural visual search task; preliminary results indicate substantial agreement between eye movements predicted by the model and those recorded from human subjects. ======================================================================== Retrieval information: FTP-host: ftp.cs.rochester.edu FTP-pathname: /pub/u/rao/papers/nips95.ps.Z URL: ftp://ftp.cs.rochester.edu/pub/u/rao/papers/nips95.ps.Z WWW URL: http://www.cs.rochester.edu:80/u/rao/ 7 pages; 570K compressed e-mail: rao@cs.rochester.edu =========================================================================