From hinton@cs.toronto.edu Tue Apr 1 13:48:59 1997 Received: from lucy.cs.wisc.edu (lucy.cs.wisc.edu [128.105.2.11]) by sea.cs.wisc.edu (8.6.12/8.6.12) with ESMTP id NAA17264 for ; Tue, 1 Apr 1997 13:48:53 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU (TELNET-1.SRV.CS.CMU.EDU [128.2.254.108]) by lucy.cs.wisc.edu (8.7.6/8.7.3) with SMTP id NAA09704 for ; Tue, 1 Apr 1997 13:48:52 -0600 (CST) Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id aa10830; 1 Apr 97 13:04:56 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id aa10825; 1 Apr 97 12:48:08 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa11508; 1 Apr 97 12:47:28 EST Received: from CS.CMU.EDU by B.GP.CS.CMU.EDU id aa25099; 1 Apr 97 12:05:25 EST Received: from yonge.cs.toronto.edu by CS.CMU.EDU id aa12420; 1 Apr 97 12:03:36 EST Received: from neuron.ai.toronto.edu ([128.100.3.14]) by yonge.cs.toronto.edu with SMTP id <86536-8370>; Tue, 1 Apr 1997 12:03:26 -0500 Received: from localhost by neuron.ai.toronto.edu with SMTP id <809>; Tue, 1 Apr 1997 12:03:21 -0500 To: Mailing List Subject: new paper available Date: Tue, 1 Apr 1997 12:03:17 -0500 From: Geoffrey Hinton Message-Id: <97Apr1.120321edt.809@neuron.ai.toronto.edu> "Generative Models for Discovering Sparse Distributed Representations" Geoffrey E. Hinton and Zoubin Ghahramani Department of Computer Science University of Toronto ABSTRACT We describe a hierarchical, generative model that can be viewed as a non-linear generalization of factor analysis and can be implemented in a neural network. The model uses bottom-up, top-down and lateral connections to perform Bayesian perceptual inference correctly. Once perceptual inference has been performed the connection strengths can be updated using a very simple learning rule that only requires locally available information. We demonstrate that the network learns to extract sparse, distributed, hierarchical representations. The paper is available at http://www.cs.toronto.edu/~hinton/ftp/RGBN.ps.Z From qian@brahms.cpmc.columbia.edu Tue Apr 1 19:59:45 1997 Received: from lucy.cs.wisc.edu (lucy.cs.wisc.edu [128.105.2.11]) by sea.cs.wisc.edu (8.6.12/8.6.12) with ESMTP id TAA19359 for ; Tue, 1 Apr 1997 19:59:40 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU (TELNET-1.SRV.CS.CMU.EDU [128.2.254.108]) by lucy.cs.wisc.edu (8.7.6/8.7.3) with SMTP id TAA16049 for ; Tue, 1 Apr 1997 19:59:39 -0600 (CST) Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id aa11288; 1 Apr 97 19:45:37 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id aa11286; 1 Apr 97 19:36:47 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa11636; 1 Apr 97 19:36:20 EST Received: from EDRC.CMU.EDU by B.GP.CS.CMU.EDU id aa28747; 1 Apr 97 15:06:04 EST Received: from brahms.cpmc.columbia.edu by EDRC.CMU.EDU id aa04050; 1 Apr 97 15:05:09 EST Received: (from qian@localhost) by brahms.cpmc.columbia.edu (8.7.6/8.7.3) id PAA19974; Tue, 1 Apr 1997 15:07:33 -0700 Date: Tue, 1 Apr 1997 15:07:33 -0700 Message-Id: <199704012207.PAA19974@brahms.cpmc.columbia.edu> From: Ning Qian To: Connectionists@cs.cmu.edu Subject: vision postdoc position at Columbia Reply-to: qian@brahms.cpmc.columbia.edu Postdoctoral Position in Visual Psychophysics Center for Neurobiology and Behavior Columbia University New York, NY A postdoctoral fellowship position in Visual Psychophysics is available immediately in my lab at Columbia University. The individual will participate in psychophysics projects that investigate the mechanisms of motion detection, stereoscopic depth perception and motion-stereo integration. There will be close interactions between these projects and the related computational modeling projects in the same lab. The details of our research interests and publications can be found at the web site listed below. The funding for the position is available for two years with the possibility of renewal. Applicants should have a strong background in visual psychophysics and should be able to write programs (or adapt our current psychophysics software package) for generating visual stimuli. Please send a CV, representative publications and two letters of recommendations to: Dr. Ning Qian Center for Neurobiology and Behavior Columbia University 722 W. 168th St., A730 New York, NY 10032 qian@brahms.cpmc.columbia.edu (email) 212-960-2213 (phone) 212-960-2561 (fax) I will attend the ARVO meeting in May. The phone number of my hotel is 954-525-3484. Please email me if you would like to arrange a meeting there. *********************************************************************h The details of our research interests and publications can be found at: http://brahms.cpmc.columbia.edu Selected Papers: Binocular Disparity and the Perception of Depth [Review], Ning Qian, Neuron, 1997, 18:359-368. The Effect of Complex Motion Pattern on Speed Perception, Bard J Geesaman and Ning Qian (submitted to Vision Research). A Novel Speed Illusion Involving Expansion and Rotation Patterns, Bard J Geesaman and Ning Qian, Vision Research, 1996, 36:3281-3292. Transparent Motion Perception as Detection of Unbalanced Motion Signals I: Psychophysics, Ning Qian, Richard A. Andersen and Edward H. Adelson, J. Neurosci., 1994, 14:7357--7366. A Physiological Model for Motion-stereo Integration and a Unified Explanation of the Pulfrich-like Phenomena, Ning Qian and Richard A. Andersen, Vision Research, 1997 (in press). Physiological Computation of Binocular Disparity, Ning Qian and Yudong Zhu, Vision Research, 1997 (in press). Binocular Receptive Field Profiles, Disparity Tuning and Characteristic Disparity, Yudong Zhu and Ning Qian, Neural Computation, 1996, 8:1647-1677. Computing Stereo Disparity and Motion with Known Binocular Cell Properties, Ning Qian, Neural Computation, 1994, 6:390-404. From cas-cns@cns.bu.edu Wed Apr 2 18:44:38 1997 Received: from lucy.cs.wisc.edu (lucy.cs.wisc.edu [128.105.2.11]) by sea.cs.wisc.edu (8.6.12/8.6.12) with ESMTP id SAA01588 for ; Wed, 2 Apr 1997 18:44:28 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU (TELNET-1.SRV.CS.CMU.EDU [128.2.254.108]) by lucy.cs.wisc.edu (8.7.6/8.7.3) with SMTP id SAA05663 for ; Wed, 2 Apr 1997 18:44:27 -0600 (CST) Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id aa12487; 2 Apr 97 17:24:05 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id aa12445; 2 Apr 97 16:49:42 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa12449; 2 Apr 97 16:48:46 EST Received: from MAILBOX.SRV.CS.CMU.EDU by B.GP.CS.CMU.EDU id aa23914; 1 Apr 97 10:51:55 EST Received: from MAILBOX.SRV.CS.CMU.EDU by MAILBOX.SRV.CS.CMU.EDU id aa12402; 1 Apr 97 10:50:52 EST Received: from CNS.BU.EDU by MAILBOX.SRV.CS.CMU.EDU id aa12399; 1 Apr 97 10:50:26 EST Received: from seer by cns.bu.edu (8.8.5/BU-11/2/94) id KAA26707; Tue, 1 Apr 1997 10:47:29 -0500 (EST) Message-Id: <3.0.1.32.19970401104517.006b2534@cns.bu.edu> X-Sender: cas-cns@cns.bu.edu (Unverified) X-Mailer: Windows Eudora Pro Version 3.0.1 (32) Date: Tue, 01 Apr 1997 10:45:17 -0500 From: BU - Cognitive and Neural Systems Subject: VISION, RECOGNITION, ACTION: FINAL CALL Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" **** FINAL CALL FOR REGISTRATION ***** International Conference on VISION, RECOGNITION, ACTION: NEURAL MODELS OF MIND AND MACHINE May 28--31, 1997 Sponsored by the Center for Adaptive Systems and the Department of Cognitive and Neural Systems Boston University with financial support from the Defense Advanced Research Projects Agency and the Office of Naval Research This conference will include 21 invited lectures and 88 contributed lectures and posters by experts on the biology and technology of how the brain and other intelligent systems see, understand, and act upon a changing world. The program is listed below. Since seating at the meeting is limited, early registration is recommended. To register, please fill out the registration form below. Student registrations must be accompanied by a letter of verification from a department chairperson or faculty/research advisor. If paying by check, mail to: Neural Models of Mind and Machine, c/o Cynthia Bradford, Boston University, Department of Cognitive and Neural Systems, 677 Beacon Street, Boston, MA 02215. If paying by credit card, mail to the above address, or fax to (617) 353-7755. The meeting registration fee will help to pay for a reception, 6 coffee breaks, and the meeting proceedings. A day of tutorials will be held on Wednesday, May 28. The tutorial registration fee helps to pay for 2 coffee breaks and a hard copy of the 7 hours of tutorial viewgraphs. See the meeting web page at http://cns-web.bu.edu/cns-meeting for further meeting information. **************************************** REGISTRATION FORM (Please Type or Print) Vision, Recognition, Action: Neural Models of Mind and Machine Boston University Boston, Massachusetts Tutorials: May 28, 1997 Meeting: May 29-31, 1997 Mr/Ms/Dr/Prof: Name: Affiliation: Address: City, State, Postal Code: Phone and Fax: Email: The conference registration fee includes the meeting program, reception, coffee breaks, and meeting proceedings. For registered participants in the conference, the regular tutorial registration fee is $20 and the student fee is $15. For attendees of only the tutorial, the regular registration fee is $30 and the student fee is $25. Two coffee breaks and a tutorial handout will be covered by the tutorial registration fee. CHECK ONE: [ ] $55 Conference plus Tutorial (Regular) [ ] $40 Conference plus Tutorial (Student) [ ] $35 Conference Only (Regular) [ ] $25 Conference Only (Student) [ ] $30 Tutorial Only (Regular) [ ] $25 Tutorial Only (Student) Method of Payment: [ ] Enclosed is a check made payable to "Boston University". Checks must be made payable in US dollars and issued by a US correspondent bank. Each registrant is responsible for any and all bank charges. [ ] I wish to pay my fees by credit card (MasterCard, Visa, or Discover Card only). Type of card: Name as it appears on the card: Account number: Expiration date: Signature and date: **************************************** MEETING SCHEDULE (poster session details follow the oral session schedule) WEDNESDAY, MAY 28, 1997 (Tutorials): 7:30am---8:30am MEETING REGISTRATION 8:30am--10:00am Stephen Grossberg (Part I): "Vision, Brain, and Technology" 10:00am--10:30am COFFEE BREAK 10:30am--12:00pm Stephen Grossberg (Part II): "Vision, Brain, and Technology" 12:00pm---1:15pm LUNCH 1:15pm---3:15pm Gail Carpenter: "Self-Organizing Neural Networks for Learning, Recognition, and Prediction: ART Architectures and Applications" 3:15pm---3:45pm COFFEE BREAK 3:45pm---5:45pm Eric Schwartz: "Algorithms and Hardware for the Application of Space-Variant Active Vision to High Performance Machine Vision" THURSDAY, MAY 29, 1997 (Invited Lectures and Posters): 7:30am---8:30am MEETING REGISTRATION 8:30am---9:15am Robert Shapley: "Brain Mechanisms for Visual Perception of Occlusion" 9:15am--10:00am George Sperling: "An Integrated Theory for Attentional Processes in Vision, Recognition, and Memory" 10:00am--10:30am COFFEE BREAK AND POSTER SESSION I 10:30am--11:15am Patrick Cavanagh: "Direct Recognition" 11:15am--12:00pm Stephen Grossberg: "Perceptual Grouping during Neural Form and Motion Processing" 12:00pm---1:30pm LUNCH 1:30pm---2:15pm Robert Desimone: "Neuronal Mechanisms of Visual Attention" 2:15pm---3:00pm Ennio Mingolla: "Visual Search" 3:00pm---3:30pm COFFEE BREAK AND POSTER SESSION I 3:30pm---4:15pm Patricia Goldman-Rakic: "The Machinery of Mind: Models from Neurobiology" 4:15pm---5:00pm Larry Squire: "Brain Systems for Recognition Memory" 5:00pm---8:00pm POSTER SESSION I FRIDAY, MAY 30, 1997 (Invited and Contributed Lectures): 8:00am---8:30am MEETING REGISTRATION 8:30am---9:15am Lance Optican: "Neural Control of Rapid Eye Movements" 9:15am--10:00am John Kalaska: "Reaching to Visual Targets: Cerebral Cortical Neuronal Mechanisms" 10:00am--10:30am COFFEE BREAK 10:30am--11:15am Rodney Brooks: "Models of Vision-Based Human Interaction" 11:15am--12:00pm Alex Pentland: "Interpretation of Human Action" 12:00pm---1:30pm LUNCH 1:30pm---1:45pm Paolo Gaudiano: "Retinal Processing of IRFPA Imagery" 1:45pm---2:00pm Zili Liu: "2D Ideal Observers in 3D Object Recognition" 2:00pm---2:15pm Soheil Shams: "Object Segmentation and Recognition via a Network of Resonating Spiking Neurons" 2:15pm---2:30pm Wey-Shiuan Hwang and John Weng: "Autonomous Learning for Visual Attention Selection" 2:30pm---2:45pm Shane W. McWhorter, Theodore J. Doll, and Anthony A. Wasilewski: "Integration of Computational Vision Research Models for Visual Performance Prediction" 2:45pm---3:00pm Frank S. Holman III and Robert J. Marks II: "Platform Independent Geometry Verification Using Neural Networks Including Color Visualization" 3:00pm---3:30pm COFFEE BREAK 3:30pm---3:45pm Heiko Neumann and Wolfgang Sepp: "A Model of Cortico-Cortical Integration of Visual Information: Receptive Fields, Grouping, and Illusory Contours" 3:45pm---4:00pm Constance S. Royden: "A Biological Model for Computing Observer Motion in the Presence of Moving Objects" 4:00pm---4:15pm Michele Fabre-Thorpe, Ghislaine Richard, and Simon Thorpe: "Rapid Categorization of Natural Images in Rhesus Monkeys: Implications for Models of Visual Processing" 4:15pm---4:30pm Raju S. Bapi and Michael J. Denham: "Neural Network Model of Experiments on Set-Shifting Paradigm" 4:30pm---4:45pm Jose L. Contreras-Vidal and George E. Stelmach: "Adaptive Resonance Theory Computations in the Cortico-Striatal Circuits are Gated by Dopamine Activity during Reward-Related Learning of Approach Behavior" 4:45pm---5:00pm Mingui Sun, Murat Sonmez, Ching-Chung Li, and Robert J. Sclabassi: "Application of Time-Frequency Analysis, Artificial Neural Networks, and Decision Making Theory to Localization of Electrical Sources in the Brain Based on Multichannel EEG" 5:00pm---6:30pm MEETING RECEPTION 6:30pm---7:30pm Stuart Anstis Keynote Lecture: "Moving in Unexpected Directions" SATURDAY, MAY 31 (Invited Lectures and Posters): 8:00am---8:30am MEETING REGISTRATION 8:30am---9:15am Eric Schwartz: "Multi-Scale Vortex of the Brain: Anatomy as Architecture in Biological and Machine Vision" 9:15am--10:00am Terrence Boult: "Polarization Vision" 10:00am--10:30am COFFEE BREAK AND POSTER SESSION II 10:30am--11:15am Allen Waxman: "Opponent Color Models of Visible/IR Fusion for Color Night Vision" 11:15am--12:00pm Gail Carpenter: "Distributed Learning, Recognition, and Prediction in ART and ARTMAP Networks" 12:00pm---1:30pm LUNCH 1:30pm---2:15pm Tomaso Poggio: "Representing Images for Visual Learning" 2:15pm---3:00pm Michael Jordan: "Graphical Models, Neural Networks, and Variational Approximations" 3:00pm---3:30pm COFFEE BREAK AND POSTER SESSION II 3:30pm---4:15pm Andreas Andreou: "Mixed Analog/Digital Neuromorphic VLSI for Sensory Systems" 4:15pm---5:00pm Takeo Kanade: "Computational VLSI Sensors: Integrating Sensing and Processing" 5:00pm---8:00pm POSTER SESSION II POSTER SESSION I: Thursday, May 29, 1997 All posters will be displayed for the full day. Biological Vision Session: #1 Vlad Cardei, Brian Funt, and Kobus Barnard: "Modeling Color Constancy with Neural Networks" #2 E.J. Pauwels, P. Fiddelaers, and L. Van Gool: "Send in the DOGs: Robust Clustering using Center-Surround Receptive Fields" #3 Tony Vladusich and Jack Broerse: "Neural Networks for Adaptive Compensation of Ocular Chromatic Aberration and Discounting Variable Illumination" #4 Alexander Dimitrov and Jack D. Cowan: "Objects and Texture Need Different Cortical Representations" #5 Miguel Las-Heras, Jordi Saludes, and Josep Amat: "Adaptive Analysis of Singular Points Correspondence in Stereo Images" #6 Neil Middleton: "Properties of Receptive Fields in Radial Basis Function (RBF) Networks" #7 David Enke and Cihan Dagli: "Modeling the Bidirectional Interactions within and between the LGN and Area V1 Cells" #8 Scott Oddo, Jacob Beck, and Ennio Mingolla: "Texture Segregation in Chromatic Element-Arrangement Patterns" #9 David Alexander and Phil Sheridan: "Local from Global Geometry of Layers 2, 3, and 4C of the Macaque Striate Cortex" #10 Phil Sheridan and David Alexander: "Invariant Transformations on a Space-Variant Hexagonal Grid" #11 Irak Vicarte Mayer and Haruhisa Takahashi: "Simultaneous Edge Detection and Image Segmentation using Neural Networks and Color Theory" #12 Adam Reeves and Shuang Wu: "Visual Adaptation: Stochastic or Deterministic?" #13 Peter Kalocsai and Irving Biederman: "Biologically Inspired Recognition Model with Extension Fields" #14 Stephane J.M. Rainville, Frederick A.A. Kingdom, and Anthony Hayes: "Effects of Local Phase Structure on Motion Perception" #15 Alex Harner and Paolo Gaudiano: "A Neural Model of Attentive Visual Search" #16 Lars Liden, Ennio Mingolla, and Takeo Watanabe: "The Effects of Spatial Frequency, Contrast, Disparity, and Phase on Motion Integration between Different Areas of Visual Space" #17 Brett R. Fajen, Nam-Gyoon Kim, and Michael T. Turvey: "Robustness of Heading Perception Along Curvilinear Paths" #18 L.N. Podladchikova, I.A. Rybak, V.I. Gusakova, N.A. Shevtsova, and A.V. Golovan: "A Behavioral Model of Active Visual Perception" #19 Julie Epelboim and Patrick Suppes: "Models of Eye Movements during Geometrical Problem Solving" Biological Learning and Recognition Session: #20 George J. Kalarickal and Jonathan A. Marshall: "Visual Classical Rearing and Synaptic Plasticity: Comparison of EXIN and BCM Learning Rules" #21 Jean-Daniel Kant and Daniel S. Levine: "ARTCRITIC: An Adaptive Critic Model for Decision Making in Context" #22 L. Andrew Coward: "Electronic Simulation of Unguided Learning, Associative Memory, Dreaming, and Internally Generated Succession of Mental Images" #23 K. Torii, T. Kitsukawa, S. Kunifuji, and T. Matsuzawa: "A Synaptic Model by Temporal Coding" #24 Gabriel Robles-de-la-Torre and Robert Sekuler: "Learning a Virtual Object's Dynamics: Spectral Analysis of Human Subjects' Internal Representation" #25 Sheila R. Cooke, Robert Sekuler, Brendan Kitts, and Maja Mataric: "Delayed and Real-Time Imitation of Complex Visual `Gestures' " #26 Brendan Kitts, Sheila R. Cooke, Maja Mataric, and Robert Sekuler: "Improved Pattern Recognition by Combining Invariance Methods" #27 Gregory R. Mulhauser: "Can ART Dynamics Create a 'Centre of Cognitive Action' Capable of Supporting Phenomenal Consciousness?" #28 Bruce F. Katz: "The Pleasingness of Polygons" #29 Stephen L. Thaler: "Device for the Autonomous Generation of Useful Information" Control and Robotics Session: #30 John Demeris and Gillian Hayes: "Integrating Visual Perception and Action in a Robot Model of Imitation" #31 Danil V. Prokhorov and Donald C. Wunsch II: "A General Training Procedure for Stable Control with Adaptive Critic Designs" #32 Juan Cires and Pedro J. Zufiria: "Space Perception through a Self-Organizing Map for Mobile Robot Control" #33 Alex Guazzelli and Michael A. Arbib: "NeWG: The Neural World Graph" #34 Minh-Chinh Nguyen: "Robot Vision Without Calibration" #35 Erol Sahin and Paolo Gaudiano: "Real-Time Object Localization from Monocular Camera Motion" #36 Carolina Chang and Paolo Gaudiano: "A Neural Network for Obstacle Avoidance in Mobile Robots" #37 P. Gaussier, J.-P. Banquet, C. Joulain, A. Revel, and S. Zrehen: "Validation of a Hippocampal Model on a Mobile Robot" #38 J.-P. Banquet, P. Gaussier, C. Joulain, and A. Revel: "Learning, Recognition, and Generation of Tempero-Spatial Sequences by a Cortico-Hippocampal System: A Neural Network Model" POSTER SESSION II: Saturday, May 31, 1997 All posters will be displayed for the full day. Machine Vision Session: #1 Tyler C. Folsom: "Edge Detection by Sparse Sampling with Steerable Quadrature Filters" #2 Mario Aguilar and Allen M. Waxman: "Comparison of Opponent-Color Neural Processing and Principal Components Analysis in the Fusion of Visible and Thermal IR Imagery" #3 Magnus Snorrason: "A Multi-Resolution Feature Integration Model for the Next-Look Problem" #4 Charles B. Owen: "Application of Multiple Media Stream Correlation to Functional Imaging of the Brain" Machine Learning Session: #5 Mukund Balasubramanian and Stephen Grossberg: "A Neural Architecture for Recognizing 3-D Objects from Multiple 2-D Views" #6 Maartje E.J. Raijmakers and Peter C.M. Molenaar: "Exact ART: A Complete Implementation of an ART Network" #7 Danil V. Prokhorov and Lee A. Feldkamp: "On the Relationship between Derivative Adaptive Critics and Backpropagation through Time" #8 Tulay Yildirim and John S. Marsland: "Optimization by Back Propagation of Error in Conic Section Functions" #9 John M. Zachary, Jacob Barhen, Nageswara S. Rao, and Sitharama S. Iyengar: "A Dynamical Systems Approach to Neural Network Learning from Finite Examples" #10 Christos Orovas and James Austin: "Cellular Associative Neural Networks" #11 M.A. Grudin, P.J.G. Lisboa, and D.M. Harvey: "A Sparse Representation of Human Faces for Recognition" #12 Mike Y.W. Leung and David K.Y. Chiu: "Feature Selection for Two-Dimensional Shape Discrimination using Feedforward Neural Networks" #13 Robert Alan Brown: "The Creation of Order in a Self-Learning Duplex Network" #14 C.H. Chen: "Designing a Neural Network to Predict Human Responses" #15 Jean-Marc Fellous, Laurenz Wiskott, Norbert Kruger, and Christoph von der Malsburg: "Face Recognition by Elastic Bunch Graph Matching" #16 Gerard J. Rinkus: "A Monolithic Distributed Representation Supporting Multi-Scale Spatio-Temporal Pattern Recognition" #17 Harald Ruda and Magnus Snorrason: "Evaluating Automatically Constructed Hierarchies of Self-Organized Neural Network Classifiers" #18 Ken J. Tomita: "A Method for Building an Artificial Neural Network with 2/3 Dimensional Visualization of Input Data" #19 Fernando J. Corbacho and Michael A. Arbib: "Towards a Coherence Theory of the Brain and Adaptive Systems" #20 Gail A. Carpenter, Mark A. Rubin, and William W. Streilein: "ARTMAP-FD: Familiarity Discrimination of Radar Range Profiles" #21 James R. Williamson: "Multifield ARTMAP: A Network for Local, Incremental, Constructive Learning" #22 Marcos M. Campos: "Constructing Adaptive Orthogonal Wavelet Bases with Self-Organizing Feature Maps" #23 Sucharita Gopal, Curtis E. Woodcock, and Alan H. Strahler: "Fuzzy ARTMAP Classification of Global Land Cover from AVHRR Data Set" #24 A.F. Rocha and A. Serapiao: "Fuzzy Modeling of the Visual Mind" #25 Eun-Jin Kim and Yillbyung Lee: "Handwritten Hangul Recognition Based on Psychologically Motivated Model" #26 Jayadeva: "A Nonlinear Programming Based Approach to the Traveling Salesman Problem" #27 Haruhisa Takahashi: "Biologically Plausible Efficient Learning Via Local Delta Rule" #28 Raonak Zaman and Donald C. Wunsch II: "Prediction of Yarn Strength from Fiber Properties using Fuzzy ARTMAP" VLSI Session: #29 James Waskiewicz and Gert Cauwenberghs: "The Boundary Contour System on a Single Chip: Analog VLSI Architecture" #30 Marc Cohen, Pamela Abshire, and Gert Cauwenberghs: "Current Mode VLSI Fuzzy ART Processor with On-Chip Learning" #31 Shinji Karasawa, Senri Ikeda, Yong Hea Ku, and Jun Hum Chung: "Methodology of the Decision-Making Device" #32 Todd Hinck and Allyn E. Hubbard: "Circuits that Implement Shunting Neurons and Steerable Spatial Filters" Audition, Speech, and Language Session: #33 Colin Davis and Sally Andrews: "Competitive and Cooperative Effects of Similarity in Stationary and Self-Organizing Models of Visual Word Recognition" #34 Susan L. McCabe and Michael J. Denham: "Towards a Neurocomputational Model of Auditory Perception" #35 Dave Johnson: "A Wavelet-Based Auditory Planning Space for Production of Vowel Sounds" #36 Michael A. Cohen, Stephen Grossberg, and Christopher Myers: "A Neural Model of Context Effects in Variable-Rate Speech Perception" #37 Peter Cariani: "Neural Computation in the Time Domain" #38 N.K. Kasabov and R. Kozma: "Chaotic Adaptive Fuzzy Neural Networks and their Applications for Phoneme-Based Spoken Language Recognition" **************************************** MEETING HOTEL INFORMATION: For all hotels listed below, meeting attendees should make their own reservations directly with the hotel using the meeting name "Vision, Recognition, Action". 1. THE ELIOT HOTEL 370 Commonwealth Avenue Boston, MA 02215 (617) 267-1607 (800) 443-5468 Janet Brown, director of sales $130/night is the Boston University rate, and is the lowest rate that the Eliot will offer to anyone, whether individual or group. A block of 12 rooms is being held until April 28, 1997. This hotel is 3 or 4 blocks from the CNS Department. 2. HOWARD JOHNSONS 575 Commonwealth Avenue Boston, MA 02215 (617) 267-3100 (reservations) (617) 864-0300 (sales office) Eric Perryman, group sales office Rates: $115/night/single and $125/night/double. A block of 15 rooms is being held until April 28, 1997. This hotel is across the street from the CNS Department. 3. THE BUCKMINSTER 645 Beacon Street Boston, MA 02215 (617) 236-7050 (800) 727-2825 Dan Betro, group sales office A block of 29 rooms is being held until April 28, 1997. This hotel is a few steps away from the CNS Department. Pricing will vary depending on the kind of room; the range is $55/night up to $129/night. Please inquire directly with the hotel when making your reservation. 4. HOLIDAY INN, BROOKLINE 1200 Beacon Street Brookline, MA 02146 (617) 277-1200 (800) 465-4329 Lisa Pedulla, Director of Sales, x-320 $99/night single, $109/night double are the Boston University rates. A block of 25 rooms will be held for us until April 28, 1997. This hotel is within a mile of the CNS Department. There is a trolley stop directly outside the hotel that will take you to within a block of the CNS Department. For information about other Boston-area hotels, please see http://www.boston.com/travel/lodging.htm. From john@dcs.rhbnc.ac.uk Wed Apr 2 18:46:30 1997 Received: from lucy.cs.wisc.edu (lucy.cs.wisc.edu [128.105.2.11]) by sea.cs.wisc.edu (8.6.12/8.6.12) with ESMTP id SAA01586 for ; Wed, 2 Apr 1997 18:44:27 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU (TELNET-1.SRV.CS.CMU.EDU [128.2.254.108]) by lucy.cs.wisc.edu (8.7.6/8.7.3) with SMTP id SAA05658 for ; Wed, 2 Apr 1997 18:44:25 -0600 (CST) Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id aa12443; 2 Apr 97 17:14:43 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id aa12441; 2 Apr 97 16:48:47 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa12438; 2 Apr 97 16:47:34 EST Received: from CS.CMU.EDU by B.GP.CS.CMU.EDU id aa11440; 2 Apr 97 5:08:10 EST Received: from [134.219.44.52] by CS.CMU.EDU id aa21242; 2 Apr 97 5:07:40 EST Received: from localhost (localhost [127.0.0.1]) by platon.cs.rhbnc.ac.uk (8.6.9/8.6.9) with SMTP id KAA31415 ; Wed, 2 Apr 1997 10:56:14 +0100 From: John Shawe-Taylor Message-Id: <199704020956.KAA31415@platon.cs.rhbnc.ac.uk> X-Authentication-Warning: platon.cs.rhbnc.ac.uk: Host localhost didn't use HELO protocol To: john@dcs.rhbnc.ac.uk, vovk@dcs.rhbnc.ac.uk, alex@dcs.rhbnc.ac.uk, pete@dcs.rhbnc.ac.uk, dave@dcs.rhbnc.ac.uk, jon@dcs.rhbnc.ac.uk, m.anthony@lse.ac.uk, Paul.Vitanyi@cwi.nl, Esko Ukkonen , orponen@igi.tu-graz.ac.at, Michel.Cosnard@lip.ens-lyon.fr, maass@igi.tu-graz.ac.at, cesabian@dsi.unimi.it, mauri , vigna@dsi.unimi.it, Felipe Cucker , Veronique.Bruyere@umh.ac.be, Christian.Michaux@umh.ac.be, Maurice.Boffa@umh.ac.be, meer@rwth-aachen.de, gavalda@lsi.upc.es, balqui@lsi.upc.es, torras@ic.upc.es, bruf@igi.tu-graz.ac.at, jpd@pip.fpms.ac.be, panizza@cse.ucsc.edu, ferretti@dsi.unimi.it, g.r.brightwell@lse.ac.uk, hpaugam@lip.ens-lyon.fr, mschmitt@igi.tu-graz.ac.at, boldi@dsi.unimi.it, Bernard.Girau@lip.ens-lyon.fr, Tapio.Elomaa@cs.Helsinki.FI, jkivinen@varisluoto.cs.Helsinki.FI, Petri.Myllymaki@cs.Helsinki.FI, koiran@lip.ens-lyon.fr, pauer@igi.tu-graz.ac.at, Didier.Puzenat@lip.ens-lyon.fr, Richard.Baron@lip.ens-lyon.fr, castro@lsi.upc.es, buhrman@cwi.nl, david@goliat.upc.es, vlavin@lsi.upc.es, carlos@goliat.upc.es, pdg@cwi.nl, N.L.Biggs@lse.ac.uk, Herman.Ehrenburg@cwi.nl, gegout@clipper.ens.fr, Olivier.Bournez@lip.ens-lyon.fr, Jean-Sylvestre.Gakwaya@umh.ac.be, simon@nereus.informatik.uni-dortmund.de, shai@csa.CS.Technion.AC.IL, bartlett@deakin.anu.edu.au, williams@faceng.anu.edu.au, lugosi@upf.es, colt@cs.uiuc.edu, Connectionists@cs.cmu.edu, neur-sci@dl.ac.uk, comp-neuro@smaug.bbb.caltech.edu, neuron-request@cattell.psych.upenn.edu Subject: Technical Report Series in Neural and Computational Learning Date: Wed, 02 Apr 97 10:56:14 +0100 X-Mts: smtp The European Community ESPRIT Working Group in Neural and Computational Learning Theory (NeuroCOLT) has produced a set of new Technical Reports available from the remote ftp site described below. They cover topics in real valued complexity theory, computational learning theory, and analysis of the computational power of continuous neural networks. Abstracts are included for the titles. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-015: ---------------------------------------- Exact Learning of subclasses of CDNF formulas with membership queries by Carlos Domingo, Universitat Polit\`ecnica de Catalunya, Spain Abstract: We consider the exact learnability of subclasses of Boolean formulas from membership queries alone. We show how to combine known learning algorithms that use membership and equivalence queries to obtain new learning results only with memberships. In particular we show the exact learnability of read-$k$ monotone CDNF formulas, Sat-$k$ ${\cal O}(\log n)$-CDNF, and ${\cal O}(\sqrt{\log n})\mbox{-size CDNF}$ from membership queries only. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-016: ---------------------------------------- Decision Trees have Approximate Fingerprints by Victor Lavin, Universitat Polit\`ecnica de Catalunya, Spain Vijay Raghavan, Vanderbilt University, USA Abstract: We prove that decision trees exhibit the ``approximate fingerprint'' property, and therefore are not polynomially learnable using only equivalence queries. A slight modification of the proof extends this result to several other representation classes of boolean concepts which have been studied in computational learning theory. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-017: ---------------------------------------- Learning Monotone Term Decision Lists by David Guijarro, Victor Lavin, Universitat Polit\`ecnica de Catalunya, Spain Vijay Raghavan, Vanderbilt University, USA Abstract: We study the learnability of monotone term decision lists in the exact model of equivalence and membership queries. We show that, for any constant $k \ge 0$, $k$-term monotone decision lists are exactly and properly learnable with $n^{O(k)}$ membership queries in O($n^{k^3}$) time. We also show $n^{\Omega (k)}$ membership queries are necessary for exact learning. In contrast, both $k$-term monotone decision lists ($k \ge 2$) and general monotone decision lists are not learnable with equivalence queries alone. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-018: ---------------------------------------- Learning nearly monotone $k$-term DNF by Jorge Castro, David Guijarro, Victor Lavin, Universitat Polit\`ecnica de Catalunya, Spain Abstract: This note studies the learnability of the class $k$-term DNF with a bounded number of negations per term. We study the case of learning with membership queries alone, and give tight upper and lower bounds on the number of negations that makes the learning task feasible. We also prove a negative result for equivalence queries. Finally, we show that a slight modification in our algorithm proves that the considered class is also learnable in the Simple PAC model, extending Li and Vit\'anyi result for monotone $k$-term DNF. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-019: ---------------------------------------- $\delta$-uniform BSS Machines by Paolo Boldi, Sebastiano Vigna, Universit\`a degli Studi di Milano, Italy Abstract: A $\delta$-uniform BSS machine is almost like a standard BSS machine, but the negativity test is replaced by a ``smaller than $-\delta$'' test, where the threshold $\delta\in(0,1)$ is not known: in this way we represent the impossibility of performing exact equality tests. We prove that, for any real closed archimedean field $R$, the $\delta$-uniform semi-decidable sets are exactly the interiors of BSS semi-decidable sets. Then, we show that the sets semi-decidable by Turing machines are the sets semi-decidable by $\delta$-uniform machines with coefficients in $Q$ or $T$, the field of Turing computable numbers. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-020: ---------------------------------------- The Computational Power of Spiking Neurons Depends on the Shape of the Postsynaptic Potentials by Wolfgang Maass, Berthold Ruf, Technische Universitaet Graz, Austria Abstract: Recently one has started to investigate the computational power of spiking neurons (also called ``integrate and fire neurons''). These are neuron models that are substantially more realistic from the biological point of view than the ones which are traditionally employed in artificial neural nets. It has turned out that the computational power of networks of spiking neurons is quite large. In particular they have the ability to communicate and manipulate analog variables in spatio-temporal coding, i.e.~encoded in the time points when specific neurons ``fire'' (and thus send a ``spike'' to other neurons). These preceding results have motivated the question which details of the firing mechanism of spiking neurons are essential for their computational power, and which details are ``accidental'' aspects of their realization in biological ``wetware''. Obviously this question becomes important if one wants to capture some of the advantages of computing and learning with spatio-temporal coding in a new generation of artificial neural nets, such as for example pulse stream VLSI. The firing mechanism of spiking neurons is defined in terms of their postsynaptic potentials or ``response functions'', which describe the change in their electric membrane potential as a result of the firing of another neuron. We consider in this article the case where the response functions of spiking neurons are assumed to be of the mathematically most elementary type: they are assumed to be step-functions (i.e. piecewise constant functions). This happens to be the functional form which has so far been adapted most frequently in pulse stream VLSI as the form of potential changes (``pulses'') that mimic the role of postsynaptic potentials in biological neural systems. We prove the rather surprising result that in models without noise the computational power of networks of spiking neurons with arbitrary piecewise constant response functions is strictly weaker than that of networks where the response functions of neurons also contain short segments where they increase respectively decrease in a linear fashion (which is in fact biologically more realistic). More precisely we show for example that an addition of analog numbers is impossible for a network of spiking neurons with piecewise constant response functions (with any bounded number of computation steps, i.e. spikes), whereas addition of analog numbers is easy if the response functions have linearly increasing segments. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-021: ---------------------------------------- On the Effect of Analog Noise in Discrete-Time Analog Computations by Wolfgang Maass, Technische Universitaet Graz, Austria Pekka Orponen, University of Jyv\"askyl\"a, Finland Abstract: We introduce a model for analog noise in analog computation with discrete time that is flexible enough to cover the most important concrete cases, such as noisy analog neural nets and networks of spiking neurons. We show that the presence of arbitrarily small amounts of analog noise reduces the power of analog computational models to that of finite automata, and we also prove a new type of upper bound for the VC-dimension of computational models with analog noise. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-022: ---------------------------------------- Networks of Spiking Neurons Can Emulate Arbitrary Hopfield Nets in Temporal Coding by Wolfgang Maass and Thomas Natschl"ager, Technische Universitaet Graz, Austria Abstract: A theoretical model for analog computation in networks of spiking neurons with temporal coding is introduced and tested through simulations in GENESIS. It turns out that the use of multiple synapses yields very noise robust mechanisms for analog computations via the timing of single spikes in networks of detailed compartmental neuron models. One arrives in this way at a method for emulating arbitrary Hopfield nets with spiking neurons in temporal coding, yielding new models for associative recall of spatio-temporal firing patterns. We also show that it suffices to store these patterns in the efficacies of \emph{excitatory} synapses. A corresponding \emph{layered} architecture yields a refinement of the synfire-chain model that can assume a fairly large set of different stable firing patterns for different inputs. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-023: ---------------------------------------- The Perceptron algorithm vs. Winnow: linear vs. logarithmic mistake bounds when few input variables are relevant by Jyrki Kivinen, University of Helsinki, Finland Manfred Warmuth, University of California, Santa Cruz, USA Peter Auer, Technische Universitaet Graz, Austria Abstract: We give an adversary strategy that forces the Perceptron algorithm to make $\Omega(k N)$ mistakes in learning monotone disjunctions over $N$ variables with at most $k$ literals. In contrast, Littlestone's algorithm Winnow makes at most $O(k\log N)$ mistakes for the same problem. Both algorithms use thresholded linear functions as their hypotheses. However, Winnow does multiplicative updates to its weight vector instead of the additive updates of the Perceptron algorithm. The Perceptron algorithm is an example of {\em additive\/} algorithms, which have the property that their weight vector is always a sum of a fixed initial weight vector and some linear combination of already seen instances. We show that an adversary can force any additive algorithm to make $(N+k-1)/2$ mistakes in learning a monotone disjunction of at most $k$ literals. Simple experiments show that for $k\ll N$, Winnow clearly outperforms the Perceptron algorithm also on nonadversarial random data. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-024: ---------------------------------------- Approximating Hyper-Rectangles: Learning and Pseudo-random Sets by Peter Auer, Technische Universitaet Graz, Austria Philip Long, National University of Singapore, Singapore Aravind Srinivasan, National University of Singapore, Singapore Abstract: The PAC learning of rectangles has been studied because they have been found experimentally to yield excellent hypotheses for several applied learning problems. Also, pseudorandom sets for rectangles have been actively studied recently because (i) they are a subproblem common to the derandomization of depth-2 (DNF) circuits and derandomizing Randomized Logspace, and (ii) they approximate the distribution of $n$ independent multivalued random variables. We present improved upper bounds for a class of such problems of ``approximating'' high-dimensional rectangles that arise in PAC learning and pseudorandomness. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-025: ---------------------------------------- On Learning from Multi-Instance Examples: Empirical Evaluation of a Theoretical Approach by Peter Auer, Technische Universitaet Graz, Austria Abstract: We describe a practical algorithm for learning axis-parallel high-dimensional boxes from multi-instance examples. The first solution to this practical learning problem arising in drug design was given by Dietterich, Lathrop, and Lozano-Perez. A theoretical analysis was performed by Auer, Long, Srinivasan, and Tan. In this work we derive a competitive algorithm from theoretical considerations which is completely different from the approach taken by Dietterich et. al. Our algorithm uses for learning only simple statistics of the training data and avoids potentially hard computational problems which were solved by heuristics by Dietterich et. al. In empirical experiments our algorithm performs quite well although it does not reach the performance of the fine-tuned algorithm of Dietterich et. al. We conjecture that our approach can be fruitfully applied also to other learning problems where certain statistical assumptions are satisfied. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-026: ---------------------------------------- Computing Functions with Spiking Neurons in Temporal Coding by Berthold Ruf, Technische Universitaet Graz, Austria Abstract: For fast neural computations within the brain it is very likely that the timing of single firing events is relevant. Recently Maass has shown that under certain weak assumptions functions can be computed in temporal coding by leaky integrate-and-fire neurons. Here we demonstrate with the help of computer simulations using GENESIS that biologically more realistic neurons can compute linear functions in a natural and straightforward way based on the basic principles of the construction given by Maass. One only has to assume that a neuron receives all its inputs in a time intervall of approximately the length of the rising segment of its excitatory postsynaptic potentials. We also show that under certain assumptions there exists within this construction some type of activation function being computed by such neurons, which allows the fast computation of arbitrary continuous bounded functions. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-027: ---------------------------------------- Hebbian Learning in Networks of Spiking Neurons Using Temporal Coding by Berthold Ruf, Michael Schmitt, Technische Universitaet Graz, Austria Abstract: Computational tasks in biological systems that require short response times can be implemented in a straightforward way by networks of spiking neurons that encode analogue values in temporal coding. We investigate the question how spiking neurons can learn on the basis of differences between firing times. In particular, we provide learning rules of the Hebbian type in terms of single spiking events of the pre- and postsynaptic neuron and show that the weights approach some value given by the difference between pre- and postsynaptic firing times with arbitrary high precision. Our learning rules give rise to a straightforward possibility for realizing very fast pattern analysis tasks with spiking neurons. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-028: ---------------------------------------- Overview of Learning Systems produced by NeuroCOLT Partners by NeuroCOLT Partners Abstract: This NeuroCOLT Technical Report documents a number of systems that have been produced withing the NeuroCOLT partnership. It only includes a summary of each system together with pointers to where the system is located and more information about its performance and design can be found. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-029: ---------------------------------------- On Bayesian Case Matching by Petri Kontkanen, Petri Myllym\"aki, Tom Silander and Henry Tirri, University of Helsinki, Finland Abstract: In this paper we present a new probabilist formalization of the case-based reasoning paradigm. In contrast to earlier Bayesian approaches, the new formalization does not need a transformation step between the original case space and the distribution space. We concentrate on applying this Bayesian framework to the case matching problem, and propose a probabilistic scoring metric for this task. In the experimental part of the paper, the Bayesian case matching score is evaluated empirically by using publicly available real-world case bases. The results show that when encountered with cases where some of the feature values have been removed, a relatively small number of remaining values is sufficient for retrieving the original case from the case base by using the proposed measure. The experiments also show that the approach is computationally very efficient. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-030: ---------------------------------------- Batch Classifications with Discrete Finite Mixtures by Petri Kontkanen, Petri Myllym\"aki, Tom Silander and Henry Tirri, University of Helsinki, Finland Abstract: In this paper we study batch classification problems where multiple predictions can be made simultaneously, instead of performing the classifications independently one at a time. For the predictions we use the model family of discrete finite mixtures, where, by introducing a hidden latent variable, we implicitly assume missing data that has to be estimated in order to be able to construct models from sample data. The main contribution of this paper is to demonstrate how the standard EM algorithm can be modified for estimating both the missing latent variable data, and the batch classification data at the same time, thus allowing us to use the same algorithm both for constructing the models from training data and for making predictions. In our framework the amount of data available for making predictions is greater than with the traditional approach, as the algorithm can also exploit the information available in the query vectors. In the empirical part of the paper, the results obtained by the batch classification approach are compared to those obtained by standard (independent) predictions by using public domain classification data sets. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-031: ---------------------------------------- Bayes Optimal Lazy Learning by Petri Kontkanen, Petri Myllym\"aki, Tom Silander and Henry Tirri, University of Helsinki, Finland Abstract: In this paper we present a new probabilistic formalization of the lazy learning approach. In our Bayesian framework, moving from the construction of an explicit hypothesis to a lazy learning approach, where predictions are made by combining the training data at query time, is equivalent to integrating out all the model parameters. Hence in Bayesian Lazy Learning the predictions are made by using all the (infinitely many) models. We present the formalization of this general framework, and illustrate its use in practice in the case of the Naive Bayes classifier model family. The Bayesian lazy learning approach is validated empriically with public domain data sets and the results are compared to the performance of the traditional, single model Naive Bayes. The general framework described in this paper can be applied with any formal model family, and to any discrete prediction task where the number of simultaneously predicted attributes is small, which includes for example all classification tasks prevalent in the machine learning literature. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-032: ---------------------------------------- On Predictive Distributions and Bayesian Networks by Petri Kontkanen, Petri Myllym\"aki, Tom Silander and Henry Tirri, University of Helsinki, Finland Abstract: In this paper we are interested in discrete prediction problems for a decision-theoretic setting, where the task is to compute the predictive distribution for a finite set of possible alternatives. This question is first addressed in a general framework, where we consider a set of probability distributions defined by some parametric model class. The standard Bayesian approach is to compute the posterior probability for the model parameters, given a prior distribution and sample data, and fix the parameters to the instantiation with the {\em maximum a posteriori} probability. A more accurate predictive distribution can be obtained by comupting the {\em evidence}, i.e., the integral over all the individual parameter instantiations. As an alternative to these two approaches, we demonstrate how to use Rissanen's new definition of {\em stochastic complexity} for determining predictive distributions. We then describe how these predictive inference methods can be realized in the case of Bayesian networks. In particular, we demonstrate the use of Jeffrey's prior as the prior distribution for computing the evidence predictive distribution. It can be shown that the evidence predictive distribution with Jeffrey's prior approaches the new stochastic complexity predictive distribution in the limit with increasing amount of sample data. For computational reasons in the experimental part of the paper the three predictive distributions are compared by using the tree-structures simple Naive Bayes model. The experimentation with several public domain classification datasets suggest that the evidence approach produces the most accurate predictions in the log-score sense, especially with small training sets. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-033: ---------------------------------------- Partial Occam's Razor and its Applications by Carlos Domingo, Tatsuie Tsukiji and Osamu Watanabe, Tokyo Institute of Technology, Japan Abstract: We introduce the notion of ``partial Occam algorithm''. A partial Occam algorithm produces a succinct hypothesis that is partially consistent with given examples, where the proportion of consistent examples is a bit more than half. By using this new notion, we propose one approach for obtaining a PAC learning algorithm. First, as shown in this paper, a partial Occam algorithm is equivalent to a weak PAC learning algorithm. Then by using boosting techniques of Schapire or Freund, we can obtain an ordinary PAC learning algorithm from this weak PAC learning algorithm. We demonstrate with some examples that some improvement is possible by this approach, in particular in the hypothesis size. First, we obtain a (non-proper) PAC learning algorithm for $k$-DNF, which has similar sample complexity as Littlestone's Winnow, but produces hypothesis of size polynomial in $d$ and $\log k$ for a $k$-DNF target with $n$ variables and $d$ terms ({\it Cf.}~ The hypothesis size of Winnow is $\CO(n^k)$). Next we show that 1-decision lists of length $d$ with $n$ variables are (non-proper) PAC learnable by using $\dsp{\CO\rpr{\frac{1}{\epsilon} \rpr{\log \frac{1}{\delta}+16^d\log n(d+\log \log n)^2}}}$ examples within polynomial time w.r.t.\ $n$, $2^d$, $1/\epsilon$, and $\log 1/\delta$. Again, we obtain a sample complexity similar to Winnow for the same problem but with a much smaller hypothesis size. We also show that our algorithms are robust against random classification noise. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-034: ---------------------------------------- Algorithms for Learning Finite Automata from Queries: A Unified View by Jos\'e Balc\'azar, Josep D\'iaz, Ricard Gavalda Universitat Polit\`ecnica de Catalunya, Spain Osamu Watanabe, Tokyo Institute of Technology, Japan Abstract: In this survey we compare several known variants of the algorithm for learning deterministic finite automata via membership and equivalence queries. We believe that our presentation makes it easier to understand what is going on and what the differences between the various algorithms mean. We also include the comparative analysis of the algorithms, review some known lower bounds, prove a new one, and discuss the question of parallelizing this sort of algorithms. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-035: ---------------------------------------- Using Fewer Examples to Simulate Equivalence Queries by Ricard Gavalda, Universitat Polit\`ecnica de Catalunya, Spain Abstract: It is well known that an algorithm that learns exactly using Equivalence queries can be transformed into a PAC algorithm that asks for random labelled examples. The first transformation due to Angluin (1988) uses a number of examples quadratic in the number of queries. Later, Littlestone (1989) and Schuurmans and Greiner (1995) gave transformations using linearly many examples. We present here another analysis of Littlestone's transformation which is both simpler and gives better leading constants. Our constants are still worse than Schuurmans and Greiner's, but while ours is a worst-case bound on the number of examples to achieve PAC learning, theirs is only an expected one. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-036: ---------------------------------------- A Dichotomy Theorem for Learning Quantified Boolean Formulas by Victor Dalmau, Universitat Polit\`ecnica de Catalunya, Spain Abstract: We consider the following classes of quantified boolean formulas. Fix a finite set of basic boolean functions. Take conjunctions of these basic functions applied to variables and constants in arbitrary way. Finally quantify existentially or universally some of the variables. We prove the following {\em dichotomy theorem}: For any set of basic boolean functions, the resulting set of formulas is either polynomially learnable from equivalence queries alone or else it is not PAC-predictable even with membership queries under cryptographic assumptions. Furthermore we identify precisely which sets of basic functions are in which of the two cases. ---------------------------------------- NeuroCOLT Technical Report NC-TR-97-037: ---------------------------------------- Discontinuities in Recurrent Neural Networks by Ricard Gavald\`a, Universitat Polit\`ecnica de Catalunya, Spain Hava Siegelmann, Technion, Israel Abstract: This paper studies the computational power of various discontinuous real computational models that are based on the classical analog recurrent neural network (ARNN). This ARNN consists of finite number of neurons; each neuron computes a polynomial net-function and a sigmoid-like continuous activation-function. The authors introduce ``arithmetic networks'' as ARNN augmented with a few simple discontinuous (eg., threshold) neurons. They argue that even with weights restricted to polynomial time computable reals, arithmetic networks are able to compute arbitrary complex recursive functions. A proof is provided to show that arithmetic networks are computationally equivalent to networks comprised of neurons that compute divisions and polynomials net-functions inside sigmoid-like continuous activation functions. Further, the authors prove that these arithmetic networks are equivalent to the Blum-Shub-Smale (BSS) model, when the latter is restricted to a bounded number of registers. With regards to implementation on digital computers, the authors demonstrate that arithmetic networks with rational weights require exponential precision; but even with very simple real weights arithmetic networks are not subject to precision bounds. As such, they can not be approximated on digital machines. This is in contrast with the ARNN that are known to demand only precision that is linear in the computation time. When complex periodic discontinuous neurons (eg. sine, tangent, fractional parts) are augmented to arithmetic networks, the resulting networks are computationally equivalent to a massively parallel machine. Thus, this highly discontinuous network can solve the presumably intractable class of PSPACE-complete problems in polynomial time. -------------------------------------------------------------------- ***************** ACCESS INSTRUCTIONS ****************** The Report NC-TR-97-001 can be accessed and printed as follows % ftp ftp.dcs.rhbnc.ac.uk (134.219.96.1) Name: anonymous password: your full email address ftp> cd pub/neurocolt/tech_reports ftp> binary ftp> get nc-tr-97-001.ps.Z ftp> bye % zcat nc-tr-97-001.ps.Z | lpr -l Similarly for the other technical reports. Uncompressed versions of the postscript files have also been left for anyone not having an uncompress facility. In some cases there are two files available, for example, nc-tr-97-002-title.ps.Z nc-tr-97-002-body.ps.Z The first contains the title page while the second contains the body of the report. The single command, ftp> mget nc-tr-97-002* will prompt you for the files you require. A full list of the currently available Technical Reports in the Series is held in a file `abstracts' in the same directory. The files may also be accessed via WWW starting from the NeuroCOLT homepage: http://www.dcs.rhbnc.ac.uk/research/compint/neurocolt or directly to the archive: ftp://ftp.dcs.rhbnc.ac.uk/pub/neurocolt/tech_reports Best wishes John Shawe-Taylor From krogh@frame.cbs.dtu.dk Thu Apr 3 14:59:58 1997 Received: from lucy.cs.wisc.edu (lucy.cs.wisc.edu [128.105.2.11]) by sea.cs.wisc.edu (8.6.12/8.6.12) with ESMTP id OAA09627 for ; Thu, 3 Apr 1997 14:59:44 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU (TELNET-1.SRV.CS.CMU.EDU [128.2.254.108]) by lucy.cs.wisc.edu (8.7.6/8.7.3) with SMTP id OAA22944 for ; Thu, 3 Apr 1997 14:59:41 -0600 (CST) Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id aa13375; 3 Apr 97 4:31:11 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id aa13373; 3 Apr 97 4:20:57 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa13161; 3 Apr 97 4:20:27 EST Received: from RI.CMU.EDU by B.GP.CS.CMU.EDU id aa29413; 3 Apr 97 2:24:34 EST Received: from [130.225.67.207] by RI.CMU.EDU id aa22118; 3 Apr 97 2:24:01 EST Received: by frame.cbs.dtu.dk (931110.SGI/930416.SGI.AUTO) for Connectionists@cs.cmu.edu id AA24093; Thu, 3 Apr 97 09:25:04 +0200 From: Anders Krogh Message-Id: <9704030924.ZM24091@frame.cbs.dtu.dk> Date: Thu, 3 Apr 1997 09:24:46 -0600 X-Mailer: Z-Mail (3.1.0 22feb94 MediaMail) To: Connectionists@cs.cmu.edu Subject: Papers on ensemble learning Cc: pkso@holyrood.ed.ac.uk Content-Type: text/plain; charset=us-ascii Mime-Version: 1.0 Dear connectionists, the following paper is now available from our web page (http://www.ph.ed.ac.uk/~pkso/papers/EnsemblePREVII.ps.gz) : STATISTICAL MECHANICS OF ENSEMBLE LEARNING Anders Krogh and Peter Sollich (Physical Review E, 55:811-825, 1997) Abstract Within the context of learning a rule from examples, we study the general characteristics of learning with ensembles. The generalization performance achieved by a simple model ensemble of linear students is calculated exactly in the thermodynamic limit of a large number of input components and shows a surprisingly rich behavior. Our main findings are the following. For learning in large ensembles, it is advantageous to use underregularized students, which actually overfit the training data. Globally optimal generalization performance can be obtained by choosing the training set sizes of the students optimally. For smaller ensembles, optimization of the ensemble weights can yield significant improvements in ensemble generalization performance, in particular if the individual students are subject to noise in the training process. Choosing students with a wide range of regularization parameters makes this improvement robust against changes in the unknown level of corruption of the training data. An abbreviated version of this paper appeared in NIPS 8 and is also available (http://www.ph.ed.ac.uk/~pkso/papers/EnsembleNIPSVI.ps.gz). Both papers are also available from http://www.cbs.dtu.dk/krogh/refs.html. Further papers of potential interest to readers of connectionists can be found on our home pages: * Peter Sollich (http://www.ph.ed.ac.uk/~pkso/publications): learning from queries, online learning, finite size effects in neural networks, ensemble learning * Anders Krogh (http://www.cbs.dtu.dk/krogh/): Most current work on using hidden Markov models in `computational biology.' All comments and suggestions are welcome - Anders Krogh and Peter Sollich -------------------------------------------------------------------------- Peter Sollich Department of Physics University of Edinburgh e-mail: P.Sollich@ed.ac.uk Kings Buildings phone: +44 - (0)131 - 650 5293 Mayfield Road fax: +44 - (0)131 - 650 5212 Edinburgh EH9 3JZ, U.K. -------------------------------------------------------------------------- Anders Krogh Center for Biological Sequence Analysis (CBS) Technical University of Denmark Building 206 DK-2800 Lyngby DENMARK Phone: +45 4525 2470 Fax: +45 4593 4808 E-mail: krogh@cbs.dtu.dk _____________________________________________ From nburgess@lbs.ac.uk Thu Apr 3 14:59:59 1997 Received: from lucy.cs.wisc.edu (lucy.cs.wisc.edu [128.105.2.11]) by sea.cs.wisc.edu (8.6.12/8.6.12) with ESMTP id OAA09629 for ; Thu, 3 Apr 1997 14:59:46 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU (TELNET-1.SRV.CS.CMU.EDU [128.2.254.108]) by lucy.cs.wisc.edu (8.7.6/8.7.3) with SMTP id OAA22946 for ; Thu, 3 Apr 1997 14:59:44 -0600 (CST) Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id ab13375; 3 Apr 97 4:31:49 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id ab13373; 3 Apr 97 4:20:59 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa13166; 3 Apr 97 4:20:36 EST Received: from CS.CMU.EDU by B.GP.CS.CMU.EDU id aa01177; 3 Apr 97 4:19:21 EST Received: from [163.119.254.19] by CS.CMU.EDU id aa02447; 3 Apr 97 4:18:56 EST Received: from neptune.lbs.ac.uk (neptune [163.119.254.10]) by galileo.lbs.ac.uk (8.7.5/8.7.3) with ESMTP id LAA06318 for ; Thu, 3 Apr 1997 11:08:43 +0100 Received: from NEPTUNE/SpoolDir by neptune.lbs.ac.uk (Mercury 1.21); 3 Apr 97 09:16:47 GMT+0 Received: from SpoolDir by NEPTUNE (Mercury 1.21); 3 Apr 97 09:16:41 GMT+0 From: Neil Burgess Organization: London Business School To: Connectionists@cs.cmu.edu Date: Thu, 3 Apr 1997 09:16:34 BST Subject: Pre-prints available - Neural Networks in the Capital Markets Reply-to: boguntula@lbs.ac.uk Priority: normal X-mailer: Pegasus Mail for Windows (v2.52) Message-ID: <9B468B447F1@neptune.lbs.ac.uk> Neural Networks in the Capital Markets: The following NNCM-96 pre-prints are now available on request. Please send your postal address to: boguntula@lbs.ac.uk ===================================================== ASSET ALLOCATION ACROSS EUROPEAN EQUITY INDICES USING A PORTFOLIO OF DYNAMIC COINTEGRATION MODELS A. N. BURGESS Department of Decision Science London Business School Regents Park, London, NW1 4SA, UK In modelling financial time-series, the model selection process is complicated by the presence of noise and possible structural non-stationarity. Additionally the near-efficiency of financial markets combined with the flexibility of advanced modelling techniques creates a significant risk of "data-snooping". These factors combine to make trading a single model a very risky proposition, particularly in a situation which allows for high leverage, such as futures trading. We believe that the risks inherent in relying on a given model can be reduced by combining a whole set of models and, to this end, describe a population-based methodology which involves building a portfolio of complementary models. We describe an application of the technique to the problem of modelling a set of European equity indices using a portfolio of cointegration-based models. ===================================================== FORECASTING VOLATILITY MISPRICING P. J. BOLLAND & A. N. BURGESS Department of Decision Science London Business School Regents Park, London, NW1 4SA, UK A simple strategy is employed to exploit volatility mispricing based on discrepancies between implied and actual market volatility. The strategy uses forward and Log contracts to either buy or sell volatility depending on whether volatility is over or under priced. As expected, buying volatility gives small profits on average but with occasional large losses in adverse market conditions. In this paper multivariate non-linear methods are used to forecast the returns of a Log contract portfolio. The explanatory power of implied volatility and the volatility term structure from several indices (FTSE, CAC, DAX) are investigated. Neural network methodologies are benchmarked against linear regression. The use of both multivariate data and non-linear techniques are shown to significantly improve the accuracy of predictions. Keywords: Options, Volatility Mispricing, Log contract, Volatility Term Structure, ===================================================== From w.penny@ic.ac.uk Thu Apr 3 23:10:29 1997 Received: from lucy.cs.wisc.edu (lucy.cs.wisc.edu [128.105.2.11]) by sea.cs.wisc.edu (8.6.12/8.6.12) with ESMTP id XAA13218 for ; Thu, 3 Apr 1997 23:10:23 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU (TELNET-1.SRV.CS.CMU.EDU [128.2.254.108]) by lucy.cs.wisc.edu (8.7.6/8.7.3) with SMTP id XAA00216 for ; Thu, 3 Apr 1997 23:10:21 -0600 (CST) Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id aa14184; 3 Apr 97 18:03:15 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id aa14182; 3 Apr 97 17:43:05 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa13595; 3 Apr 97 17:42:34 EST Received: from RI.CMU.EDU by B.GP.CS.CMU.EDU id aa06063; 3 Apr 97 9:18:04 EST Received: from romeo.ic.ac.uk by RI.CMU.EDU id aa23778; 3 Apr 97 9:17:28 EST Received: from judy.ic.ac.uk [155.198.5.5] by romeo.ic.ac.uk with esmtp (Exim 0.57 #1) id 0wCnJq-0007Z9-00; Thu, 3 Apr 1997 15:16:50 +0100 Received: from rosalind.ee.ic.ac.uk (rosalind.ee.ic.ac.uk [155.198.139.10]) by judy.ic.ac.uk (8.7.5/8.7.5) with SMTP id PAA16203; Thu, 3 Apr 1997 15:16:21 +0100 (BST) Received: from albert.ee.ic.ac.uk by rosalind.ee.ic.ac.uk (4.1/4.1) id AA03241; Thu, 3 Apr 97 15:08:27 BST From: w.penny@ic.ac.uk Date: Thu, 3 Apr 1997 15:22:20 +0100 Message-Id: <22983.199704031422@albert.ee.ic.ac.uk> To: Connectionists@cs.cmu.edu Subject: Research jobs in neural nets and pattern recognition Cc: w.penny@ic.ac.uk, s.j.roberts@ic.ac.uk, rickman@rrl.co.uk X-Sun-Charset: ISO-8859-1 THREE POST-DOCTORAL RESEARCH POSITIONS IN PATTERN RECOGNITION / NEURAL NETWORKS RESEARCH Three post-doctoral research positions are available within the Neural Systems Section of the Department of Electrical & Electronic Engineering to work on the theory and application of advanced pattern recognition techniques, in particular the use of Bayesian methods and neural networks. Two positions are funded for two years and the third nominally for three years with a yearly evaluation. All projects involve research in statistical pattern recognition with applications in the biomedical field. Experience in pattern recognition and Bayesian statistics would be an advantage. A good understanding of data processing (especially signal processing) techniques is desired as is experience of UNIX, C and Matlab. The positions are funded by the Jefferiss Research Trust, the European Commission and British Aerospace plc respectively. The salary scale will be RA1A , GBP 14,732 - 22,143 per annum (exclusive of London Allowance of GBP 2,134) depending on age and experience. Further information may be obtained from http://www.ee.ic.ac.uk/research/neural/positions.html or via e-mail to Dr Stephen Roberts (s.j.roberts@ic.ac.uk). The closing date for applications is April 11th 1997. In recent years, great interest has developed in the use of non-classical methods for statistical analysis of data as part of a general increase towards the use of artificial intelligence methods. One genre which has shown itself to be particularly suitable is that of connectionist models, a subset of which are referred to as a artificial neural networks (ANNs). Classical statistical methods rely upon the use of simple models, such as linear or logistic regression, in order to 'learn' relationships between variables and outcomes. ANNs offer a far more flexible model set, indeed it has been shown that they have the property of universal approximation so they are able, in principle, to estimate any set of arbitrary relationships between variables. Furthermore, they may model non-linear coupling between sets of variables. Part of the momentum of the recent development of ANNs for pattern recognition, regression and estimation problems must be attributed to the manner in which ANNs conform to many of the traditional statistical approaches, i.e. they may estimate Bayesian probablilities in the case of classification and conditional averages in the case of regression. 1) The use of Neural Networks to Predict the Development and Progression of Kaposi's Sarcoma (KS). This is a joint project funded by the Jefferiss Research Trust between the Department of Electrical and Electronic Engineering, Imperial College of Science, Technology & Medicine and the Department of Genito-urinary Medicine, St. Mary's Hospital. Kaposi's sarcoma (KS) is a vascular tumour, which is more common and often aggressive in patients with underlying immunosuppression (post-transplant KS and AIDS-associated KS). KS was first described by the Hungarian pathologist Moritz Kaposi in 1872, yet still remains something of a clinical enigma, being an unusual tumour of unknown origin. The aim of this research is to determine factors that influence the variable progression rate of KS in HIV infected individuals. There is currently no means of predicting which patients will develop KS and no understanding of the relationship between the forms of the disease. The aim of the project is to carry out multi-variable analyses in order to define clinical end-points and provide guidelines for better patient management. A number of variables will be available to the system. The reliability and utility of each with regard to the prediction of patient outcome, however, is generally unknown. Classical regression analysis offers some powerful methods of selection and ranking within a subset of features or variables. Whilst such methods should be used for completeness and comparison, it is noted that recent developments in Bayesian learning theory have offered the possibility to assess the utility of variables from within the ANN structure. Each input variable has a separate weighting factor, or in Bayesian terminology, a hyper-prior, associated with it. This technique has become known as automatic relevance determination or ARD. Such an assessment is devoid of the strong assumptions of independence and linearity of most of the classical regression methods. It is feasible for an ANN to produce not only a set of output variables (predictions or classifications, for example) but also an associated set of confidence or validation measures (describing the probable error on each output). This enables the tracking of predictions of future events in a more robust framework and furthermore allows for the accurate fusion of information from more than one source and the incorporation of temporal information, i.e. poor quality information from the present time may be suppressed in favour of more reliable information from past or future as it becomes available. If we may regard the system as aiming to produce a probability distribution in some 'outcome space', then several possible approaches to analysis are made available. As temporal information is retained (i.e. outcomes are based upon the entire course of the patient's history, not just present information) we may seek information regarding the effect of each piece of information (test result or partial diagnosis) on the probability distribution in the 'outcome space'. Two pieces of information may be obtained from this approach. How important a partial decision or test result is to the probability of certain outcomes and how important it is to changing the uncertainty we have in the outcome results. Clearly, the goal will be to indicate tests and/or procedures which not only increase the survivability probabilities but also make the estimated outcomes less variant, so we have more confidence in the predictions (this means not only increasing the height of a favourable node in the posterior probability space, but also attempting to reduce the variance of the distribution). In order to accommodate for multiple output hypotheses we propose to utilise a procedure similar to that detailed in (Bishop 1995) whereby the output distribution is modelled multi- modally. This has the added benefit that individual modes (possible outcomes) may be tracked separately. This representation is also similar to that taken in a mixture of experts approach. REFERENCES 1. Bishop CM. Neural Networks for Pattern Recognition. Oxford University Press, Oxford, 1995. 2. Ripley BD. Pattern Recognition and Neural Networks. Cambridge University Press, Cambridge, 1996. 3. Roberts SJ and Penny W. Novelty, Confidence and Errors in Connectionist Systems. Proceedings of IEE colloquium on fault detection and intelligent sensors, IEE, September 1996. 4. Penny W and Roberts SJ. Neural Networks with Error Bars. Departmental report, also submitted to IEEE transactions on neural networks, February 1997, available from http://www.ee.ic.ac.uk/staff/hp/sroberts.html 2) SIESTA (EU funded project) Siesta is a an EU funded project which involves Imperial College and 10 other European partners. The aim of the project is to define and produce a system which is capable of continuous evaluation of the state of the brain during the sleep-wake cycle. Such an automated system is of enormous value in the clinical field and the research into multi-channel signal processing, fusion and pattern recognition form a challenge to the most modern techniques. The state of the brain will, primarily, be monitored via its electrical activity (the EEG). One of the most well-known approaches from literature to achieve a continuous description of EEG state is the system developed by Roberts & Tarassenko (1992a, 1992b). This approach will be used as a general basis for the research in SIESTA. Roberts & Tarassenko (henceforth, R&T') used a self-organizing feature map (SOM) to perform unsupervised topographic mapping of feature vectors consisting of 10 coefficients of a Kalman filter algorithm applied to the raw EEG. This self- organizing network discovered eight distinct clusters in which the brain state remained preferentially. >From the general trajectories between these halting states, R&T discovered three distinct patterns which roughly corresponded to the global brain states W (wakefulness), S (slow wave or deep sleep) and R (REM sleep). The outputs from the self-organizing feature map were subsequently mapped via a Radial Basis Function (RBF) classifier onto three outputs, trained by sections of data on which experts agreed upon the stage W, S, or R. For each input, the resulting network produced probabilities for the three global stages, describing intermediate stages as a combination of three 'mixing fractions' . This general approach, yielding a novel way of describing brain state while exploiting some of experts' knowledge in a partly supervised method, will be adopted and extended in the following ways Many features extracted from the signals will be considered. Instead of the 2-dimensional feature map used by R&T, alternative approaches will be investigated. It has been shown that combinations of other clustering and mapping methods can outperform SOMs. Moreover, since topographic mapping is only exploited for visualization, the general approach can be based on more advanced clustering techniques (e.g. techniques for non-Gaussian clustering (Roberts 1997) or Bayesian-inspired methods). in order to cope with the large number of input features to be investigated active feature selection methods will be applied. techniques for intelligent sensor fusion will be investigated. When multiple sources are combined to lead to classification results, it is not trivial to decide which are the most relevant sources at any given time, or what should happen when sources fail to provide input (e.g. because an electrode is faulty). Approaches based on the computation of running error measures can be employed here. The Imperial College group will form be a leading centre in the theory subgroup of the project. We will be active in the researching of Mixture density networks and mixtures of experts Model estimation and pre-processing Active sensor fusion Active feature and data selection Unsupervised data partitioning methods (clustering) Model comparison and validation References 1. S.J. Roberts, Parametric and Non-parametric Unsupervised Cluster Analysis. Pattern Recognition, 30 (2) ,1997. 2. S.J. Roberts and L. Tarassenko, The Analysis of the Sleep EEG using a Multi- layer Neural Network with Spatial Organisation. IEE Proceedings Part F, 139(6), 420-425, 1992a. 3. S.J. Roberts and L. Tarassenko, New Method of Automated Sleep Quantification. Medical and Biological Engineering and Computing, 30(5), 509-517, 1992b. 3) Assessment of cortical vigilance This project, funded by British Aerospaces Sowerby Research Centre at Bristol, aims to assess and predict lapses in vigilance in the human brain. Recordings of the brains electrical activity are to be recorded and analysed. The utility of a device or system which may monitor an individual's level of vigilance is clear in a range of safety-critical environments. We propose that such utility would be enhanced by a system which, as well as monitoring the present state of vigilance, made a prediction as to the likely evolution of vigilance in the near future. To perform both these tasks, i.e. a static pattern assessment and a dynamic tracking and prediction, sophisticated methods of information extraction, sensor fusion and classification/regression must be employed. Over the last decade the theory of artificial neural' networks has been pitched within the framework of advanced statistical decision theory and it is within this framework which we intend to work. The aim of the project is to work towards a practical real-time system. The latter should be minimally intrusive and should make predictions of future vigilance states. Where appropriate, therefore, the investigation will assess each technique in the developing system with a view to its implementation in a real-time environment. The project will involve research into : New methods of signal complexity and synchronisation estimation Information flow estimation in multi-channel environments Active sensor fusion Prediction and classification Error estimation State transition detection and state sequence modelling Reference Makeig, S. and Jung, T-P. and Sejnowski, T. (1996), Using feedforward neural networks to monitor alertness from changes in EEG correlation and coherence, Advances in Neural Information Processing Systems (NIPS), MIT Press, Cambridge, MA. From biehl@physik.uni-wuerzburg.de Fri Apr 4 12:14:24 1997 Received: from lucy.cs.wisc.edu (lucy.cs.wisc.edu [128.105.2.11]) by sea.cs.wisc.edu (8.6.12/8.6.12) with ESMTP id MAA19174 for ; Fri, 4 Apr 1997 12:14:19 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU (TELNET-1.SRV.CS.CMU.EDU [128.2.254.108]) by lucy.cs.wisc.edu (8.7.6/8.7.3) with SMTP id MAA09352 for ; Fri, 4 Apr 1997 12:14:18 -0600 (CST) Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id aa15180; 4 Apr 97 3:52:36 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id aa15178; 4 Apr 97 3:43:20 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa13906; 4 Apr 97 3:42:32 EST Received: from EDRC.CMU.EDU by B.GP.CS.CMU.EDU id aa25591; 4 Apr 97 3:32:30 EST Received: from wptx01.physik.uni-wuerzburg.de by EDRC.CMU.EDU id aa24801; 4 Apr 97 3:31:51 EST Received: (from smap@localhost) by wptx01.physik.uni-wuerzburg.de (8.7.5/8.7.3/TPhys_S) id KAA26639 for ; Fri, 4 Apr 1997 10:31:43 +0200 (MESZ) From: Michael Biehl Received: from wptx08.physik.uni-wuerzburg.de(132.187.40.8) by wptx01.physik.uni-wuerzburg.de (smap V1.3) id sma026629 for ; Fri Apr 4 10:31:37 1997 Received: (from biehl@localhost) by wptx08.physik.uni-wuerzburg.de (8.7.5/8.7.3/TPhys_C) id KAA29194 for Connectionists@cs.cmu.edu; Fri, 4 Apr 1997 10:31:36 +0200 (MESZ) Message-Id: <199704040831.KAA29194@wptx08.physik.uni-wuerzburg.de> Subject: preprint on on-line unsupervised learning To: Connectionists Date: Fri, 4 Apr 1997 10:31:35 +0200 (MESZ) X-Mailer: ELM [version 2.4 PL24 ME8b] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit FTP-host: ftp.physik.uni-wuerzburg.de FTP-filename: /pub/preprint/1997/WUE-ITP-97-003.ps.gz The following manuscript is now available via anonymous ftp (see below for the retrieval procedure), or, alternatively from http://www.physik.uni-wuerzburg.de/~biehl ------------------------------------------------------------------ "Specialization processes in on-line unsupervised learning" Michael Biehl, Ansgar Freking, Georg Reents, and Enno Schl"osser Contribution to the Minerva Workshop on Mesoscopics, Fractals and Neural Networks Eilat, Israel, March 1997 Ref: WUE-ITP-97-003 Abstract >From the recent analysis of supervised learning by on-line gradient descent in multilayered neural networks it is known that the necessary process of student specialization can be delayed significantly. We demonstrate that this phenomenon also occurs in various models of unsupervised learning. A solvable model of competitive learning is presented, which identifies prototype vectors suitable for the repre- sentation of high--dimensional data. The specific case of two overlapping clusters of data and a matching number of prototype vectors exhibits non- trivial behavior like almost stationary plateau configurations. As a second example scenario we investigate the application of Sanger's algorithm for principal component analysis in the presence of two relevant directions in input space. Here, the fast learning of the first principal component may lead to an almost complete loss of initial knowledge about the second one. --------------------------------------------------------------------- Retrieval procedure: unix> ftp ftp.physik.uni-wuerzburg.de Name: anonymous Password: {your e-mail address} ftp> cd pub/preprint/1997 ftp> binary ftp> get WUE-ITP-97-003.ps.gz (*) ftp> quit unix> gunzip WUE-ITP-97-003.ps.gz e.g. unix> lp WUE-ITP-97-003.ps [10 pages] (*) can be replaced by "get WUE-ITP-97-003.ps". The file will then be uncompressed before transmission (slow!). _____________________________________________________________________ -- Michael Biehl Institut fuer Theoretische Physik Julius-Maximilians-Universitaet Wuerzburg Am Hubland D-97074 Wuerzburg email: biehl@physik.uni-wuerzburg.de homepage: http://www.physik.uni-wuerzburg.de/~biehl Tel.: (+49) (0)931 888 5865 " " " 5131 Fax : (+49) (0)931 888 5141 From sami@guillotin.hut.fi Fri Apr 4 19:53:53 1997 Received: from lucy.cs.wisc.edu (lucy.cs.wisc.edu [128.105.2.11]) by sea.cs.wisc.edu (8.6.12/8.6.12) with ESMTP id TAA21423 for ; Fri, 4 Apr 1997 19:53:43 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU (TELNET-1.SRV.CS.CMU.EDU [128.2.254.108]) by lucy.cs.wisc.edu (8.7.6/8.7.3) with SMTP id TAA16858 for ; Fri, 4 Apr 1997 19:53:42 -0600 (CST) Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id aa15965; 4 Apr 97 15:10:00 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id aa15938; 4 Apr 97 14:49:13 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa14312; 4 Apr 97 14:48:27 EST Received: from EDRC.CMU.EDU by B.GP.CS.CMU.EDU id aa29770; 4 Apr 97 7:57:04 EST Received: from guillotin.hut.fi by EDRC.CMU.EDU id aa25647; 4 Apr 97 7:56:58 EST Received: (from sami@localhost) by guillotin.hut.fi (950413.SGI.8.6.12/950213.SGI.AUTOCF) id OAA01930; Fri, 4 Apr 1997 14:56:49 +0200 Date: Fri, 4 Apr 1997 14:56:49 +0200 Message-Id: <199704041256.OAA01930@guillotin.hut.fi> From: Sami Kaski To: connectionists@cs.cmu.edu Subject: Thesis on data exploration with SOMs available Reply-to: sami.kaski@hut.fi The following Dr.Tech. thesis is available at http://nucleus.hut.fi/~sami/thesis/thesis.html (html-version) http://nucleus.hut.fi/~sami/thesis.ps.gz (compressed postscript, 300K) http://nucleus.hut.fi/~sami/thesis.ps (postscript, 2M) The articles that belong to the thesis can be accessed through the page http://nucleus.hut.fi/~sami/thesis/node3.html --------------------------------------------------------------- Data Exploration Using Self-Organizing Maps Samuel Kaski Helsinki University of Technology Neural Networks Research Centre P.O.Box 2200 (Rakentajanaukio 2C) FIN-02015 HUT, Finland Finding structures in vast multidimensional data sets, be they measurement data, statistics, or textual documents, is difficult and time-consuming. Interesting, novel relations between the data items may be hidden in the data. The self-organizing map (SOM) algorithm of Kohonen can be used to aid the exploration: the structures in the data sets can be illustrated on special map displays. In this work, the methodology of using SOMs for exploratory data analysis or data mining is reviewed and developed further. The properties of the maps are compared with the properties of related methods intended for visualizing high-dimensional multivariate data sets. In a set of case studies the SOM algorithm is applied to analyzing electroencephalograms, to illustrating structures of the standard of living in the world, and to organizing full-text document collections. Measures are proposed for evaluating the quality of different types of maps in representing a given data set, and for measuring the robustness of the illustrations the maps produce. The same measures may also be used for comparing the knowledge that different maps represent. Feature extraction must in general be tailored to the application, as is done in the case studies. There exists, however, an algorithm called the adaptive-subspace self-organizing map, recently developed by Kohonen, which may be of help. It extracts invariant features automatically from a data set. The algorithm is here characterized in terms of an objective function, and demonstrated to be able to identify input patterns subject to different transformations. Moreover, it could also aid in feature exploration: the kernels that the algorithm creates to achieve invariance can be illustrated on map displays similar to those that are used for illustrating the data sets. From movellan@ergo.ucsd.edu Fri Apr 4 19:53:54 1997 Received: from lucy.cs.wisc.edu (lucy.cs.wisc.edu [128.105.2.11]) by sea.cs.wisc.edu (8.6.12/8.6.12) with ESMTP id TAA21425 for ; Fri, 4 Apr 1997 19:53:44 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU (TELNET-1.SRV.CS.CMU.EDU [128.2.254.108]) by lucy.cs.wisc.edu (8.7.6/8.7.3) with SMTP id TAA16860 for ; Fri, 4 Apr 1997 19:53:43 -0600 (CST) Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id aa16188; 4 Apr 97 18:07:13 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id aa16184; 4 Apr 97 17:48:15 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa14441; 4 Apr 97 17:47:43 EST Received: from EDRC.CMU.EDU by B.GP.CS.CMU.EDU id aa11512; 4 Apr 97 17:36:17 EST Received: from ergo.ucsd.edu by EDRC.CMU.EDU id aa29288; 4 Apr 97 17:34:58 EST Received: (from movellan@localhost) by ergo.ucsd.edu (8.8.4/8.8.4) id OAA31931; Fri, 4 Apr 1997 14:27:09 -0800 Date: Fri, 4 Apr 1997 14:27:09 -0800 Message-Id: <199704042227.OAA31931@ergo.ucsd.edu> From: "Javier R. Movellan" To: connectionists@cs.cmu.edu Subject: UCSD Cogsci TR The following technical report is available online at http://cogsci.ucsd.edu (follow links to Tech Reports & Software ) Physical copies are also available (see the site for information). Analysis of Direction Selectivity Arising From Recurrent Cortical Interactions. Paul Mineiro and David Zipser UCSD Cogsci TR.97.03 The relative contributions of feedforward and recurrent connectivity to the direction selective responses of cells in layer IVB of primary visual cortex is currently the subject of debate in the neuroscience community. Recently biophysically detailed simulations have shown realistic direction selective responses can be achieved via recurrent cortical interactions between cells with non-direction selective feedforward input \cite{Koch:DS,Maex:SpikeMotion}. Unfortunately the complexity of these models, while desirable for detailed comparison with biology, are difficult to analyze mathematically. In this paper a relatively simple cortical dynamical model is used to analyze the emergence of direction selective responses via recurrent interactions. Comparison between a model based on our analysis and physiological data is presented. The approach also allows analysis of the recurrently propagated signal, revealing the predictive nature of the implementation. From riegler@ifi.unizh.ch Sat Apr 5 04:31:16 1997 Received: from lucy.cs.wisc.edu (lucy.cs.wisc.edu [128.105.2.11]) by sea.cs.wisc.edu (8.6.12/8.6.12) with ESMTP id EAA22494 for ; Sat, 5 Apr 1997 04:31:11 -0600 Received: from TELNET-1.SRV.CS.CMU.EDU (TELNET-1.SRV.CS.CMU.EDU [128.2.254.108]) by lucy.cs.wisc.edu (8.7.6/8.7.3) with SMTP id EAA20917 for ; Sat, 5 Apr 1997 04:31:09 -0600 (CST) Received: from TELNET-1.SRV.CS.CMU.EDU by telnet-1.srv.cs.CMU.EDU id ab15965; 4 Apr 97 15:11:48 EST Received: from DST.BOLTZ.CS.CMU.EDU by TELNET-1.SRV.CS.CMU.EDU id aa15945; 4 Apr 97 14:51:37 EST Received: from DST.BOLTZ.CS.CMU.EDU by DST.BOLTZ.CS.CMU.EDU id aa14322; 4 Apr 97 14:51:14 EST Received: from CS.CMU.EDU by B.GP.CS.CMU.EDU id aa00708; 4 Apr 97 8:54:30 EST Received: from josef.ifi.unizh.ch by CS.CMU.EDU id aa09877; 4 Apr 97 8:54:03 EST Received: from mac77.ifi.unizh.ch by josef.ifi.unizh.ch with SMTP (PP) id <23501-0@josef.ifi.unizh.ch>; Fri, 4 Apr 1997 15:51:08 +0200 X-Sender: riegler@josef.ifi.unizh.ch Message-Id: Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Date: Fri, 4 Apr 1997 15:51:10 +0200 From: Alex Riegler Subject: NTCS-97 Call For Participation CALL FOR PARTICIPATION /\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/ International Workshop N E W T R E N D S I N C O G N I T I V E S C I E N C E NTCS '97 /\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/ "Does Representation need Reality?" Perspectives from Cognitive Science, Neuroscience, Epistemology, and Artificial Life Vienna, Austria, May 14 - 16, 1997 with plenary talks by: Larry Cauller, Georg Dorffner, Ernst von Glasersfeld, Stevan Harnad, Wolf Singer, and Sverre Sjoelander organized by the Austrian Society of Cognitive Science (ASoCS) =========================================================================== Latest information can be retrieved from the conference WWW-page =========================================================================== P u r p o s e ___________________________________________________________________________ The goal of this single-track conference is to investigate and discuss new approaches and movements in cognitive science in a workshop-like atmosphere. Among the topics which seem to have emerged in the last years are: embodiment of knowledge, system theoretic and computational neuroscience approaches to cognition, dynamics in recurrent neural architectures, evolutionary and artificial life approaches to cognition, and (epistemological) implications for perception and representation, constructivist concepts and the problem of knowledge representation, autopoiesis, implications for epistemology and philosophy (of science). Evidence for a failure of the traditional understanding of neural representation converges from several fields. Neuroscientific results in the last decade have shown that single cell representations with hierarchical processing towards representing units seems not the way the cortex represents environmental entities. Instead, distributed cell ensemble coding has become a popular concept for representation, both in computational and in empirical neuroscience. However, new problems arise from the new concepts. The problem of binding the distributed parts into a uniform percept can be "solved" by introducing synchronization of the member neurons. A deeper (epistemological) problem, however, is created by recurrent architectures within ensembles generating an internal dynamics in the network. The cortical response to an environmental stimulus is no longer dominated by stimulus properties themselves, but to a considerable degree by the internal state of the network. Thus, a clear and stable reference between a representational state (e.g. in a neuron, a Hebbian ensemble, an activation state, etc.) and the environmental state becomes questionable. Already learned experiences and expectancies might have an impact on the neural activity which is as strong as the stimulus itself. Since these internally stored experiences are constantly changing, the notion of (fixed) representations is challenged. At this point, system theory and constructivism, both investigating the interaction between environment and organism at an abstract level, come into the scene and turn out to provide helpful epistemological concepts. The goal of this conference is to discuss these phenomena and their implications for the understanding of representation, semantics, language, cognitive science, and artificial life. Contrary to many conferences in this field, the focus is on interdisciplinary cooperation and on conceptual and epistemological questions, rather than on technical details. We are trying to achieve this by giving more room to discussion and interaction between the participants (e.g., invited comments on papers, distribution of papers to the participants before the conference, etc.). According to the interdisciplinary character of cognitive science, we welcome papers/talks from the fields of artificial life, empirical, cognitive, and computational neuroscience, philosophy (of science), epistemology, anthropology, computer science, psychology, and linguistics. T a l k s ___________________________________________________________________________ NOTE: Names with * are invited speakers. Constructivism Ernst von Glasersfeld* Piaget's legacy: Cognition as adaptive activity Sverre Sjoelander* How animals handle reality: the adaptive aspect of representation Annika Wallin Is there a way to distinguish representation from Perception... Tom Routen Habitus and Animats General Epistemology and Methodology W.F.G. Haselager Is cognitive science advancing towards behaviorism? Michael Pauen Reality and representation William Robinson Representation and cognitive explanation Matthias Scheutz The ontological status of representations Anthony Chemero Two types of anti-representationism: a taxonomy Georg Schwarz Can representation get reality? Neuroscience Larry Cauller* NeuroInteractivism: Explaining emergence without representation Wolf Singer* The observer in the brain Steven Bressler The dynamic manifestation of cognitive structures in the cerebral cortex Erich Harth Sketchpads in and beyond the brain Marius Usher Active Neural representations: neurophysiological data and its implications Symbol Grounding and Communication Georg Dorffner* The connectionist route to embodiment and dynamicism Stevan Harnad* Keeping a grip on the real/virtual distinction in this representationalist age Mark Wexler Must mental representation be internal? Tom Ziemke Rethinking Grounding Christian Balkenius Explorations in synthetic pragmatics Horst Hendriks-Jansen Does natural cognition need internal knowledge structures? Nathan Chandler On the importance of reality in representations Peter Gaerdenfors Does semantics need reality? P o s t e r s ___________________________________________________________________________ Chris Browne Iconic Learning and Epistemology Mark Claessen RabbitWorld: the concept of space can be learned Valentin Constantinescu Interaction between perception & expectancy... Andrew Coward Unguided categorization, direct and symbolic representation, and evolution of cognition... David Davenport PAL: A constructivist model of cognitive activity Karl Diller Representation and reality: where are the rules of grammar Richard Eiser Representation and the social reality Robert French When coffe cups are like old elephants Maurita Harney Representation and its metaphysics Daniel Hutto Cognition without representation? Amy Ione Symbolic creation and re-representation of reality Sydney Lamb Top-Down Modeling, Bottom-up Learning Michael Luntley Real Representations Ralf Moeller Perception through anticipation Ken Mogi Response selectivity, neuron doctrine, and Mach's principle in perception Alfredo Pereira The term "representation" in cognitive neuroscience Michael Ramscar Judgement of association: problems with cognitive theories of analogy Hanna Risku Constructivist Consequences: does tramslation need reality? Sabine Weiss Cooperation of different neural networks during single word and sentence processing R e g i s t r a t i o n ___________________________________________________________________________ To register please fill out the registration form at the bottom of this CFP and send it by... o Email to franz-markus.peschl@univie.ac.at, or by o Fax to +43-1-408-8838 (attn. M.Peschl), or by o Mail to Markus Peschl, Dept.for Philosophy of Science (address below) Registration Fee (includes admission to talks, presentations, and proceedings): Member * 1300 ATS (about 118 US$) Non-Member 1800 ATS (about 163 US$) Student Member ** 500 ATS (about 45 US$) Student Non-Member 1300 ATS (about 118 US$) *) Members of the Austrian Society of Cognitive Science **) Requires proof of valid student ID C o n f e r e n c e S i t e a n d A c c o m o d a t i o n ___________________________________________________________________________ The conference takes place in a small beautiful baroque castle in the suburbs of Vienna; the address is: Schloss Neuwaldegg Waldegghofg. 5 A-1170 Wien Austria Tel: +43 1 485 3605 Fax: +43 1 485 3605-112 It is surrounded by a beautiful forest and a good (international and Viennese gastronomic) infrastructure. On the tram it takes only 20 minutes to the center of Vienna (see overview). (Limited) Accommodation is provided by the castle (about 41 US$ per night (single), 30 US$ per night, per person (double) including breakfast). Please contact the telephone number above. You can find more information about Vienna and accommodation at the Vienna Tourist Board or at the Intropa Travel agent Tel: +43-1-5151-242. Note: In case you want to stay over the weekend we refer you to the following hotel which is near the conference site (single about 75 US$ / 850 ATS per night): Hotel Jaeger Hernalser Hauptstrasse 187 A-1170 Wien Austria Tel: +43 1 486 6620 Fax: +43 1 486 6620 8 D e s t i n a t i o n V i e n n a ? ___________________________________________________________________________ Vienna, Austria, can be reached internationally by plane or train. The Vienna Schwechat airport is located about 16 km from the city center. From the airport, the city air-terminal can be reached by bus (ATS 60.- per person) or taxi (about ATS 400). Rail-passengers arrive at one of the main stations which are located almost in the city center. From the air-terminal and the railway stations the congress site and hotels can be reached easily by underground (U-Bahn), tramway, or bus. A detailed description will be given to the participants. In May the climate is mild in Vienna. It is the time when spring is at its climax and everything is blooming. The weather is warm with occasional (rare) showers. The temperature is about 18 to 24 degrees Celsius. More information about Vienna and Austria on the web: Welcome to Vienna Scene Vienna City Wiener Festwochen - Vienna Festival Public Transport in Vienna (subway) Welcome to Austria General information about Austria Austria Annoted S c i e n t i f i c C o m m i t t e e ___________________________________________________________________________ R. Born Univ. of Linz (A) R. Born Univ. of Linz (A) G. Dorffner Univ. of Vienna (A) E. v. Glasersfeld Univ. of Amherst, MA (USA) S. Harnad Univ. of Southampton (GB) M. Peschl Univ. of Vienna (A) A. Riegler Univ. of Zurich (CH) H. Risku Univ. of Skovde (S) M. Scheutz Univ. of Indiana (USA) W. Singer Max Planck Institut, Frankfurt (D) S. Sjoelander Linkoeping University (S) A. v. Stein Neuroscience Institute, La Jolla (USA) O r g a n i z i n g C o m m i t t e e ___________________________________________________________________________ M. Peschl Univ. of Vienna (A) A. Riegler Univ. of Zurich (CH) S p o n s o r i n g O r g a n i z a t i o n s ___________________________________________________________________________ o Christian Doppler Laboratory for Expert Systems (Vienna University of Technology) o Oesterreichische Forschgungsgemeinschaft o Austrian Federal Ministry of Science, Transport and the Arts o City of Vienna A d d i t i o n a l I n f o r m a t i o n ___________________________________________________________________________ For further information on the conference contact: Markus Peschl Dept. for Philosophy of Science University of Vienna Sensengasse 8/10 A-1090 Wien Austria Tel: +43-1-402-7601/41 Fax: +43-1-408-8838 Email: franz-markus.peschl@univie.ac.at General information about the Austrian Society for Cognitive Science can be found on the Society webpage or by contacting Alexander Riegler AILab, Dept. of Computer Science University of Zurich Winterthurerstr. 190 CH-8057 Zurich Switzerland Email: riegler@ifi.unizh.ch R e g i s t r a t i o n f o r m ___________________________________________________________________________ I participate at the Workshop "New Trends in Cognitive Science (NTCS'97)" Full Name ........................................................................ Full Postal Address: ........................................................................ ........................................................................ ........................................................................ Telephone Number (Voice): Fax: ..................................... .................................. Email address: ........................................................................ Payment in ATS (= Austrian Schillings; 1 US$ is currently about 11 ATS). This fee includes admission to talks, presentations, and proceedings: [ ] Member * 1300 ATS (about 118 US$) [ ] Non-Member 1800 ATS (about 163 US$) [ ] Student Member ** 500 ATS (about 45 US$) [ ] Student Non-Member 1300 ATS (about 118 US$) *) Members of the Austrian Society of Cognitive Science **) Requires proof of valid student ID Total: .................... ATS [ ] Visa [ ] Master-/Eurocard Name of Cardholder ........................................ Credit Card Number ........................................ Expiration Date ................. Date: ................ Signature: ........................................ Please send this form by... o Email to franz-markus.peschl@univie.ac.at, or by o Fax to +43-1-408-8838 (attn. M.Peschl), or by o Mail to Markus Peschl, Dept.for Philosophy of Science, Univ. of Vienna, Sensengasse 8/10, A-1090 Wien, Austria