ADVANCES IN NEURAL INFORMATION 
PROCESSING SYSTEMS 8 
ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 
Published by Morgan-Kaufmann 
NIPS-1 
Advances in Neural Information Processing 
Touretzky, ed., 1989 
Systems: Proceedings of the 1988 Conference, David 
NIPS-2 
Advances in Neural Information Processing 
Touretzky, ed., 1990 
Systems: Proceedings of the 1989 Conference, David 
NIPS-3 
Advances in Neural Information Processing Systems: Proceedings of the 1990 Conference, 
Richard Lippmann, John E. Moody, and David S. Touretzky, eds., 1991 
NIPS-4 
Advances in Neural Information Processing Systems: Proceedings of the 1991 Conference, John 
E. Moody, Steve J. Hanson, and Richard P. Lippmann, eds., 1992 
NIPS-5 
Advances in Neural Information Processing Systems: Proceedings of the 1992 Conference, 
Stephen Jos6 Hanson, Jack D. Cowan, and C. Lee Giles, eds., 1993 
NIPS-6 
Advances in Neural Information Processing Systems: Proceedings of the 1993 Conference, Jack 
D. Cowan, Gerald Tesauro, and Joshua Alspector, eds., 1994 
Published by The MIT Press 
NIPS-7 
Advances in Neural Information Processing Systems: Proceedings of the 1994 Conference, Gerald 
Tesauro, David Touretzky, and Todd Leen, eds., 1995 
NIPS-8 
Advances in Neural Information Processing Systems: Proceedings of the 1995 Conference, David 
S. Touretzky, Michael C. Mozer, and Michael E. Hasselmo, eds., 1996 
ADVANCES IN NEURAL INFORMATION 
PROCESSING SYSTEMS 8 
Proceedings of the 1995 Conference 
edited by 
David S. Touret:ky, Michael C. Mo:er, and Michael E. Hasselmo 
A Bradford Book 
The MIT Press 
Cambridge, Massachusetts 
London, England 
(D 1996 Massachusetts Institute of Technology 
All rights reserved. No part of this book may be reproduced in any form by any electronic or 
mechanical means (including photocopying, recording, or information storage and retrieval) with- 
out permission in writing from the publisher. 
This book was printed and bound in the United States of America. 
ISSN: 1049-5258 
ISBN: 0-262-20107-0 
Contents 
Preface 
Committees 
xvii 
PART I 
COGNITIVE SCIENCE 
Learning the Structure of Similarity 
I. B. TENENBAUM 
A Model of Spatial Representations in Parietal Cortex Explains Hemineglect 
A. POUGET, T. J. SEJNOWSKI 
Human Reading and the Curse of Dimensionality 
G. L. MARTIN 
Extracting Tree-structured Representations of Trained Networks 
M. W. CRAVEN, I. W. SHAVLIK 
Harmony Networks Do Not Work 
R. GOURLEY 
Dynamics of Attention as Near Saddle-node Bifurcation Behavior 
H. NAKAHARA, K. DOYA 
Rapid Quality Estimation of Neural Network Input Representations 
K. J. CHERKAUER, I. W. SHAVLIK 
A Model of Auditory Streaming 
S. L. MCCABE, M. I. DENHAM 
3 
10 
17 
24 
31 
38 
45 
52 
PART II 
NEUROSCIENCE 
Modeling Interactions of the Rat's Place and Head Direction Systems 
A.D. REDISH, D.S. TOURETZKY 
Correlated Neuronal Response: Time Scales and Mechanisms 
W. BAIR, E. ZOHARY, C. KOCH 
Information through a Spiking Neuron 
C. STEVENS, A. ZADOR 
Reorganization of Somatosensory Cortex after Tactile Training 
R. S. PETERSEN, I. G. TAYLOR 
61 
68 
82 
v i Contents 
A Dynamical Model of Context Dependencies for the Vestibulo-Ocular Reflex 
O. J. M.D. COENEN, T. J. SEJNOWSKI 
The Role of Activity in Synaptic Competition at the Neuromuscular Junction 
S. R. H. JOSEPH, D. J. WILLSHAW 
When Is an Integrate-and-fire Neuron like a Poisson Neuron? 
C. F. STEVENS, A. ZADOR 
How Perception Guides Production in Birdsong Learning 
C. L. FRY 
The Geometry of Eye Rotations and Listing's Law 
A. A. HANDZEL, T. FLASH 
Temporal Coding in the Submillisecond Range: Model of Barn Owl Auditory 
Pathway 
R. KEMPTER, W. GERSTNER, J. L. VAN HEMMEN, H. WAGNER 
Cholinergic Suppression of Transmission May Allow Combined Associative 
Memory Function and Self-organization in the Neocortex 
M. E. HASSELMO, M. CEKIC 
A Predictive Switching Model of Cerebellar Movement Control 
A. G. BARTO, J. T. BUCKINGHAM, J. C. HOUK 
Independent Component Analysis of Eiectroencephalographic Data 
S. MAKEIG, A. J. BELL, T.-P. JUNG, T. J. SEJNOWSKI 
Simulation of a Thalamocortical Circuit for Computing Directional Heading 
in the Rat 
H. T. BLAIR 
Plasticity of Center-Surround Opponent Receptive Fields in Real and Artificial 
Neural Systems of Vision 
S. YASUI, T. FURUKAWA, M. YAMADA, T. SAITO 
89 
96 
103 
11o 
117 
124 
131 
138 
145 
152 
159 
PART III 
THEORY 
Learning Model Bias 
J. BAXTER 
Statistical Theory of Overtraining--Is Cross-Validation Asymptotically 
Effective? 
S. AMARI, N. MURATA, K. R. Mf)LLER, M. FINKE, H. YANG 
A Bound on the Error of Cross Validation Using the Approximation and 
Estimation Rates, with Consequences for the Training-test Split 
M. KEARNS 
Learning with Ensembles: How Overfitting Can Be Useful 
P. SOLLICH, A. KROGH 
169 
176 
183 
190 
Contents vii 
Neural Networks with Quadratic VC Dimension 
P. KOIRAN, E. D. SONTAG 
Sample Complexity for Learning Recurrent Perceptron Mappings 
B. DASGUPTA, E. D. SONTAG 
On the Computational Power of Noisy Spiking Neurons 
w. MAASS 
A Realizable Learning Task Which Exhibits Overfitting 
s. BOS 
Stable Dynamic Parameter Adaptation 
S. M. ROGER 
Estimating the Bayes Risk from Sample Data 
R. R. SNAPP, T XU 
Recursive Estimation of Dynamic Modular RBF Networks 
V. KADIRKAMANATHAN, M. KADIRKAMANATHAN 
On Neural Networks with Minimal Weights 
V. BOHOSSIAN, J. BRUCK 
Modern Analytic Techniques to Solve the Dynamics of Recurrent Neural 
Networks 
A. C. C. COOLEN, S. N. LAUGHTON, D. SHERRINGTON 
Implementation Issues in the Fourier Transform Algorithm 
Y. MANSOUR, S. SAHAR 
Generalisation of a Class of Continuous Neural Networks 
J. SHAWE-TAYLOR, J. ZHAO 
Gradient and Hamiltonian Dynamics Applied to Learning in Neural Networks 
J. W. HOWSE, C. T. ABDALLAH, G. L. HEILEMAN 
Optimization Principles for the Neural Code 
M. DEWEESE 
Strong Unimodality and Exact Learning of Constant Depth g-Perceptron 
Networks 
M. MARCHAND, S. HADJIFARADJI 
Active Learning in Multilayer Perceptrons 
K. FUKUMIZU 
Dynamics of On-line Gradient Descent Learning for Multilayer Neural 
Networks 
D. SAAD, S. A. SOLLA 
Worst-case Loss Bounds for Single Neurons 
D. P. HELMBOLD, J. KIVINEN, M. K. WARMUTH 
Exponentially Many Local Minima for Single Neurons 
P. AUER, M. HERBSTIER, M. K. WARMUTH 
197 
204 
211 
218 
225 
232 
239 
246 
253 
260 
267 
274 
281 
288 
295 
302 
309 
316 
viii Contents 
Adaptive Back-Propagation in On-line Learning of Multilayer Networks 
A. H. L. WEST, D. SAAD 
Optimizing Cortical Mappings 
G. J. GOODHILL, S. FINCH, T.J. SEJNOWSKI 
Quadratic-type Lyapunov Functions for Competitive Neural Networks with 
Different Time-scales 
A. MEYER-B,SE 
Examples of Learning Curves from a Modified VC-formalism 
A. KOWALCZYK, J. SZYMANSKI, P. L. BARTLETT, R. C. WILLIAMSON 
Bayesian Methods for Mixtures of Experts 
S. WATERHOUSE, D. MACKAY, T. ROBINSON 
Some Results on Convergent Unlearning Algorithm 
S. A. SEMENOV, I. B. SHUVALOVA 
Geometry of Early Stopping in Linear Networks 
R. DODIER 
Absence of Cycles in Symmetric Neural Networks 
X. WANG, A. JAGOTA, F. BOTELHO, M. GARZON 
323 
330 
337 
344 
351 
358 
365 
372 
PART IV 
ALGORITHMS AND ARCHITECTURES 
Adaptive Mixture of Probabilistic Transducers 
Y. SINGER 
REMAP: Recursive Estimation and Maximization of A Posteriori Probabilities-- 
Application to Transition-based Connectionist Speech Recognition 
Y. KONIG, H. BOURLARD, N. MORGAN 
Recurrent Neural Networks for Missing or Asynchronous Data 
Y. BENGIO, F. GINGRAS 
Family Discovery 
S. M. OMOHUNDRO 
Discriminant Adaptive Nearest Neighbor Classification and Regression 
T. HASTIE, R. TIBSHIRANI 
Clustering Data through an Analogy to the Potts Model 
M. BLATT, S. WISEMAN, E. DOMANY 
Generalized Learning Vector Quantization 
A. SATO, K. YAMADA 
Stochastic Hillclimbing as a Baseline Method for Evaluating Genetic 
Algorithms 
A. JUELS, M. WATTENBERG 
381 
388 
395 
402 
409 
416 
423 
43O 
Contents ix 
Symplectic Nonlinear Component Analysis 
L. C. PARRA 
A Unified Learning Scheme: Bayesian-Kuilback Ying-Yang Machine 
L. XU 
Universal Approximation and Learning of Trajectories Using Oscillators 
P. BALDI, K. HORNIK 
A Smoothing Regularizer for Recurrent Neural Networks 
L. WU, J. MOODY 
EM Optimization of Latent-Variable Density Models 
C. M. BISHOP, M. SVENSIN, C. K. I. WILLIAMS 
Factorial Hidden Markov Models 
Z. GHAHRAMANI, M. I. JORDAN 
Boosting Decision Trees 
H. DRUCKER, C. CORTES 
Exploiting Tractable Substructures in Intractable Networks 
L. K. SAUL, M. I. JORDAN 
Hierarchical Recurrent Neural Networks for Long-term Dependencies 
S. E. HIHI, Y. BENGIO 
Discovering Structure in Continuous Variables Using Bayesian Networks 
R. HOFMANN, V. TRESP 
Using Pairs of Data Points to Define Splits for Decision Trees 
G. E. HINTON, M. REVOW 
Gaussian Processes for Regression 
C. K. I. WILLIAMS, C. E. RASMUSSEN 
Pruning with Generalization Based Weight Saliencies: 3BD, hOBS 
M. W. PEDERSEN, L. K. HANSEN, J. LARSEN 
Fast Learning by Bounding Likelihoods in Sigmoid Type Belief Networks 
T. JAAKKOLA, L. K. SAUL, M., I. JORDAN 
Generating Accurate and Diverse Members of a Neural-network Ensemble 
D. W. OPITZ, J. W. SHAVLIK 
Improved Gaussian Mixture Density Estimates Using Bayesian Penalty Terms 
and Network Averaging 
D. ORMONEIT, V. TRESP 
Explorations with the Dynamic Wave Model 
T. P. REBOTIER, J. L. ELMAN 
The Capacity of a Bump 
G. W. FLAKE 
Tempering Backpropagation Networks: Not All Weights Are Created Equal 
N. N. SCHRAUDOLPH, T. J. SEJNOWSKI 
437 
444 
451 
458 
465 
472 
479 
486 
493 
500 
507 
514 
521 
528 
535 
542 
549 
556 
563 
x Contents 
Investment Learning with Hierarchical PSOMs 
J. WAEYER, H. RITTER 
Learning Long-term Dependencies Is Not as Difficult with NARX Networks 
T. LIN, B. G. HORNE, P. TIl/O, C. L. GILES 
Constructive Algorithms for Hierarchical Mixtures of Experts 
S. R. WATERHOUSE, A. J. ROBINSON 
An Information-theoretic Learning Algorithm for Neural Network Classification 
D. MILLER, A. RAO, K. ROSE, A. GERSHO 
A Practical Monte Carlo Implementation of Bayesian Learning 
C. E. RASMUSSEN 
From Isolation to Cooperation: An Alternative View of a System of Experts 
S. SCHAAL, C. C. ATKESON 
Finite State Automata that Recurrent Cascade-Correlation Cannot Represent 
S. C. KREMER 
SPERT-II: A Vector Microprocessor System and Its Application to Large 
Problems in Backpropagation Training 
J. WAWRZYNEK, K. ASANOVIC, B. KINGSBURY, J. BECK, D. JOHNSON, N. MORGAN 
Softassign versus Softmax: Benchmarks in Combinatorial Optimization 
S. GOLD, A. RANGARAJAN 
A Multiscale Attentionai Framework for Relaxation Neural Networks 
D. I. TSIOUTSIAS, E. MJOLSNESS 
Is Learning the n-th Thing Any Easier Than Learning the First? 
S. THRUN 
Using Unlabeled Data for Supervised Learning 
G. TOWELL 
Learning Sparse Percepttons 
J. C. JACKSON, M. W. CRAVEN 
Does the Wake-sleep Algorithm Produce Good Density Estimators? 
B. J. FREY, G. E. HINTON, P. DAYAN 
570 
577 
584 
591 
598 
605 
612 
619 
626 
633 
640 
647 
654 
661 
PART V 
IMPLEMENTATIONS 
Improved Silicon Cochlea Using Compatible Lateral Bipolar Transistors 
A. VAN SCHAIK, E. FRAGNIRE, E. VITTOZ 
Adaptive Retina with Center-Surround Receptive Field 
S.-C. LIU, K. BOAHEN 
671 
678 
Contents x i 
Neuron-MOS Temporal Winner Search Hardware for Fully-parallel Data 
Processing 
T. SHIBATA T. NAKAI, T. MORIMOTO, R. KAIHARA, T. YAMASHITA, T. OHMI 
Analog VLSI Processor Implementing the Continuous Wavelet Transform 
R. T. EDWARDS, G. CAUWENBERGHS 
Silicon Models for Auditory Scene Analysis 
J. LAZZARO, J. WAWRZYNEK 
VLSI Model of Primate Visual Smooth Pursuit 
R. ETIENNE-CUMMINGS, J. VAN DER SPIEGEL, P. MUELLER 
Model Matching and SFMD Computation 
S. REHFUSS, D. HAMMERSTROM 
Parallel Analog VLSI Architectures for Computation of Heading Direction and 
Time-to-contact 
G. INDIVERI, J. KRAMER, C. KOCH 
685 
692 
699 
706 
713 
720 
PART VI 
SPEECH AND SIGNAL PROCESSING 
Onset-based Sound Segmentation 
L. S. SMITH 
Laterally Interconnected Self-organizing Maps in Handwritten Digit 
Recognition 
Y. CHOE, J. SIROSH, R. MIIKKULAINEN 
Forward-backward Retraining of Recurrent Neural Networks 
A. SENIOR, T. ROBINSON 
Context-dependent Classes in a Hybrid Recurrent Network-HMM Speech 
Recognition System 
D. KERSHAW, T. ROBINSON, M. HOCHBERG 
A New Learning Algorithm for Blind Signal Separation 
S. AMARI, A. CICHOCKI, H. H. YANG 
Handwritten Word Recognition Using Contextual Hybrid Radial Basis Function 
Network/Hidden Markov Models 
B. LEMARII, M. GILLOUX, M. LEROUX 
Selective Attention for Handwritten Digit Recognition 
E. ALPAYDIN 
KODAK IMAGELINK TM OCR Alphanumeric Handprint Module 
A. SHUSTOROVICH, C. W. THRASHER 
The Gamma MLP for Speech Phoneme Recognition 
S. LAWRENCE, A. C. TSOI, A.D. BACK 
729 
736 
743 
750 
757 
764 
771 
778 
785 
xii Contents 
PART VII 
VISION 
A Framework for Nonrigid Matching and Correspondence 
S. PAPPU, S. GOLD, A. RANGARAJAN 
Control of Selective Visual Attention: Modeling the "Where" Pathway 
E. NIEBUR, C. KOCH 
Unsupervised Pixel-prediction 
W. R. SOFTKY 
Learning to Predict Visibility and Invisibility from Occlusion Events 
J. A. MARSHALL, R. K. ALLEY, R. S. HUBBARD 
Classifying Facial Action 
M. S. BARTLETT, P. A. VIOLA, T. J. SEJNOWSKI, B. A. GOLOMB, J. LARSEN, 
J. C. HAGER, P. EKMAN 
Modeling Saccadic Targeting in Visual Search 
R. P. N. RAO, G. J. ZELINSKY, M. M. HAYHOE, D. H. BALLARD 
A Model of Transparent Motion and Non-transparent Motion Aftereffects 
A. GRUNEWALD 
A Neural Network Model of 3-D Lightness Perception 
L. PESSOA, W. D. ROSS 
Empirical Entropy Manipulation for Real-world Problems 
P. VIOLA, N. N. SCHRAUDOLPH T. J. SEJNOWSK! 
Active Gesture Recognition Using Learned Visual Attention 
T. DARRELL, A. PENTLAND 
SEEMORE: A View-based Approach to 3-D Object Recognition Using Multiple 
Visual Cues 
B. W. MEL 
795 
802 
809 
816 
823 
830 
837 
844 
851 
858 
865 
PART VIII 
APPLICATIONS 
Human Face Detection in Visual Scenes 
H. A. ROWLEY, S. BALUJA, T. KANADE 
Improving Committee Diagnosis with Resampling Techniques 
B. PARMANTO, P. W. MUNRO, H. R. DOYLE 
Primitive Manipulation Learning with Connectionism 
Y. MATSUOKA 
Beating a Defender in Robotic Soccer: Memory-based Learning of a 
Continuous Function 
P. STONE, M. VELOSO 
875 
882 
889 
896 
Contents xiii 
Visual Gesture-based Robot Guidance with a Modular Neural System 
E. LITTMANN, A. DREES, H. RITTER 
A Novel Channel Selection System in Cochlear Implants Using Artificial Neural 
Network 
M. A. JABRI, R. J. WANG 
Prediction of Beta Sheets in Proteins 
A. KROGH, S. K. RIIS 
A Neural Network Autoassociator for Induction Motor Failure Prediction 
T. PETSCHE, A. MARCANTONIO, C. DARKEN, S. J. HANSON, G. M. KUHN, 
I. SANTOSO 
Using Feedforward Neural Networks to Monitor Alertness from Changes in 
EEG Correlation and Coherence 
S. MAKEIG, T.-P. JUNG, T. J. SEJNOWSKI 
A Neural Network Classifier for the I1000 OCR Chip 
J. C. PLATT, T P. ALLEN 
Predictive Q-Routing: A Memory-based Reinforcement Learning Approach 
to Adaptive Traffic Control 
S. P.M. CHOI, D. YEUNG 
Optimal Asset Allocation Using Adaptive Dynamic Programming 
R. NEUNEIER 
Using the Future to "Sort Out" the Present: Rankprop and Multitask Learning 
for Medical Risk Evaluation 
R. CARUANA, S. BALUJA, T. MITCHELL 
Stock Selection via Nonlinear Multi-factor Models 
A. U. LEVIN 
Experiments with Neural Networks for Real Time Implementation of Control 
P. CAMPBELL, M. DALE, H. L. FERRY, A. KOWALCZYK 
High-speed Airborne Particle Monitoring Using Artificial Neural Networks 
A. FERGUSON, T. SABISCH, P. KAYE, L. C. DIXON, H. BOLOURI 
903 
910 
917 
924 
931 
938 
945 
952 
959 
966 
973 
980 
PART IX 
CONTROL 
A Dynamical Systems Approach for a Learnable Autonomous Robot 
J. TANI, N. FUKUMURA 
Parallel Optimization of Motion Controllers via Policy Iteration 
J. A. COELHO JR., R. SITARAMAN, R. A. GRUPEN 
Learning Fine Motion by Markov Mixtures of Experts 
M. MEILA, M. I. JORDAN 
Neural Control for Nonlinear Dynamic Systems 
S. YU, A.M. ANNASWAMY 
989 
996 
1003 
lOlO 
xiv Contents 
Improving Elevator Performance Using Reinforcement Learning 
R. H. CRITES, A. G. BARTO 
High-performance Job-Shop Scheduling with a Time-delay TD(30 Network 
W. ZHANG, T. G. DIETTERICH 
Competence Acquisition in an Autonomous Mobile Robot Using Hardware 
Neural Techniques 
G. JACKSON, A. F. MURRAY 
Generalization in Reinforcement Learning: Successful Examples Using 
Sparse Coarse Coding 
R. S. SUTTON 
Stable Linear Approximations to Dynamic Programming for Stochastic Control 
Problems with Local Transitions 
B. V. ROY, J. N. TSITSIKLIS 
Stable Fitted Reinforcement Learning 
G. J. GORDON 
Improving Policies without Measuring Merits 
P. DAYAN, S. P. SINGH 
Memory-based Stochastic Optimization 
A. W. MOORE J. SCHNEIDER 
Temporal Difference in Learning in Continuous Time and Space 
K. DOYA 
Reinforcement Learning by Probability Matching 
P. N. SABES, M. I. JORDAN 
1017 
1024 
1031 
1038 
1045 
1052 
1059 
1066 
1073 
1080 
Author Index 
Keyword Index 
1087 
1091 
Preface 
The interaction between neuroscience, cognitive science and theoretical work on information 
processing has increased rapidly over the past decade. In experimental work, the different 
levels of analysis and the complexity of the data require computational modeling in order to 
evaluate hypotheses and generate new experimentally testable predictions. At the same time, 
many information processing applications can draw inspiration from available data about the 
neural substrates of behavior. The papers in this volume contribute to this ongoing 
interaction between theory and experiment. 
These papers summarize the talks and posters presented at the ninth annual conference on 
Neural Information Processing Systems (NIPS), held in Denver, Colorado from Nov. 27 to 
Nov. 30, 1995. The previous eight volumes of Advances in Neural Information Processing 
Systems have made an influential contribution to the field of biological and artificial neural 
networks. All papers presented at the conference were reviewed by three referees, with final 
acceptance of 152 out of the 462 papers submitted. Many excellent papers could not be 
accepted due to space limitations. The acceptance rate of this volume is lower than that for 
many journals--ensuring that it contains work of only the highest quality. All submitted 
papers received reviewer comments and feedback from NIPS conference participants which 
were incorporated in the revisions for final publication in this proceedings volume. 
The field of computational neuroscience and neural network theory has drawn researchers 
from a range of more traditional disciplines. The papers in this volume describe how 
neuroscientists and cognitive scientists have used computational models of neural systems to 
test hypotheses and generate predictions to guide their work. This work includes models of 
how networks in the owl brainstem could be trained for complex localization function 
(Kempter et al.) how cellular activity may underlie rat navigation (Redish and Touretzky, 
Blair), how cholinergic modulation may regulate cortical reorganization (Hasselmo and 
Cekic), and how damage to parietal cortex results in neglect (Pouget and Sejnowski). 
Additional work concerns development of theoretical techniques important for understanding 
the dynamics of neural systems, including formation of cortical maps (Goodhill and Finch), 
analysis of recurrent networks (Coolen et al.), and analysis of self-supervised learning (Frey 
et al.). Other papers describe how engineers and computer scientists have approached 
problems of pattern recognition or speech recognition with the use of computational 
architectures inspired by the interaction of populations of neurons within the brain. For 
example, new neural network models have been applied to classical problems including 
x v i Preface 
handwritten character recognition (Lemarie et al., Choe et al) and object recognition (Mel, 
Rowley et al). Exciting new work has focused on building chips modeled after neural 
systems (Van Schaik et al., Lazzaro and Wawrzynek). 
In addition to the papers in the main program, the NIPS conference was preceded by an 
excellent one-day tutorial program, and a popular two-day program of post-conference 
workshops took place in Vail. The workshop program has grown considerably in scale over 
the years. This year it covered topics ranging from "Vertebrate Neurophysiology and Neural 
Networks" to "Learning in Bayesian Networks." 
The continued success of the NIPS conference can be attributed to the efforts of a large 
group of people, many of whom are listed on the following pages. We express our gratitude 
to the members of the organizing committee, program committee, publicity committee and 
foundation board, and the many referees who reviewed the 462 submissions to the 
conference. We especially thank Manavendra Misra for his smooth and cool handling of 
local arrangements, Christy Medina and Denise Pruell for their stellar efforts as professional 
conference administrators and registration coordinators, the many on-site student volunteers, 
John Angulo, Deb Miller, Mark Sitton, and Tim Tidwell II, students at the University of 
Colorado at Boulder who handled the monumental task of processing the conference 
submissions, and Mike Chang and Olivia Choe for assistance in assembling this volume. 
Finally, we thank the Colorado School of Mines for helping to finance the registration 
services, and the Advanced Research Projects Agency and Office of Naval Research for 
providing valuable financial support for many of the graduate students and young 
investigators who attend the meeting. 
David S. Touretzky, Carnegie Mellon University 
Michael C. Mozer, University of Colorado 
Michael E. Hasselmo, Harvard University 
January 18, 1996 
Committees 
NIPS-95 
ORGANIZING COMMITTEE 
General Chair 
Program Chair 
Workshop Chair 
Publicity Chair 
Publications Chair 
Treasurer 
Government/Corporate Liaison 
Local Arrangements Chair 
Tutorials Chair 
Contracts 
David Touretzky, Carnegie Mellon University 
Michael Mozer, University of Colorado 
Michael Perrone, IBM 
David Cohn, MIT 
Michael Hasselmo, Harvard University 
John Lazzaro, UC Berkeley 
John Moody, Oregon Graduate Institute 
Manavendra Misra, Colorado School of Mines 
Jack Cowan, University of Chicago 
Steve Hanson, Siemens 
Scott Kirkpatrick, IBM 
Gerald Tesauro, IBM 
NIPS-95 
PROGRAM COMMITTEE 
Program Chair 
Area Chairs 
Michael Mozer, University of Colorado 
Yoshua Bengio, Universite de Montreal 
Terrence Fine, Cornell University 
John G. Harris, University of Florida 
Michael Kearns, AT&T Bell Labs 
Yann Le Cun, AT&T Bell Labs 
Stephen Omohundro, NEC Research Inst. 
Lori Pratt, Colorado School of Mines 
Jose lh'incipe, University of Florida 
Allen Selverston, UC San Diego 
Satinder P. Singh, MIT 
Volker Tresp, Siemens AG 
Steven Zucker, McGill University 
xv i i i Committees 
NIPS-95 
PUBLICITY COMMITTEE 
Publicity Chair 
Overseas Liaisons 
Australia, Singapore, India 
Europe 
Hong Kong, China, Taiwan 
Israel 
Japan 
South America 
Turkey 
United Kingdom 
David Cohn, MIT 
Marwan Jabri, University of Sydney 
Joachim Buhmann, University of Bonn 
Lei Xu, Chinese University of Hong Kong 
Hava Siegelmann, Technion 
Kenji Doya, ATR Research Laboratories 
Andreas Meier, Simon Bolivar University 
Ethem Alpaydin, Bogazici University 
Alan Murray, Edinburgh University 
NIPS FOUNDATION 
BOARD MEMBERS 
President 
Vice President of Development 
Treasurer 
Secretary 
Members 
NIPS95 General Chair 
Terry Sejnowski, Salk Institute 
John Moody, Oregon Graduate Institute 
Eric Mjolsness, UC San Diego 
Scott Kirkpatrick, IBM 
Leo Breiman, UC Berkeley 
Jack Cowan, University of Chicago 
Stephen Hanson, Siemens 
Richard Lippman, MIT 
Eve Marder, Brandeis 
Gerald Tesauro, IBM 
David Touretzky, Carnegie Mellon 
NIPS-95 
REFEREES 
Subutai Ahmad 
Luis Almeida 
Chuck Anderson 
James Anderson 
Martin Anthony 
Chris Atkeson 
Pierre Baldi 
Etienne Barnard 
Andrew Barron 
Peter Bartlett 
Eric Baum 
Sue Becker 
Yoshua Bengio 
Avrim Blum 
Ulrich Bodenhausen 
Leon Bottou 
Herve Bourlard 
Jochen Braun 
Timothy Brown 
Joachim Buhmann 
John Byrne 
Franco Callaft 
Ted Camevale 
Gert Cauwenberghs 
Tzi-Dar Chiueh 
Michael Cohen 
William Cohen 
David Cohn 
Corinna Cortes 
Gary Cottrell 
Christian Darken 
Peter Dayan 
Virginia de Sa 
Bert deVries 
Thomas Dietterich 
Allan Dobbins 
Georg Dorffner 
Gregory Dudek 
Shimon Edelman 
Mark Fanty 
Meir Feder 
Terrence Fine 
Gary Flake 
Bill Freeman 
Fabrizio Gabbiani 
Patrick Gallinari 
Zoubin Ghahramani 
Joumana Ghosn 
Committees x ix 
C. Lee Giles 
Federico Girosi 
Hans Peter Graf 
Vijaykumar Gullapalli 
Patrick Haffner 
Dan Hammerstrom 
Tom Hancock 
Catherine Harris 
Michael Hasselmo 
David Haussler 
Simon Haykin 
John Hertz 
Haym Hirsh 
Bill Horue 
Don Hush 
Nathan Intrator 
Tommi J_akkolla 
Marwan Jabri 
Robbie Jacobs 
Allan Jepson 
Dan Johnston 
Leslie Pack 
Michael Kearns 
Dan Kersten 
Christof Koch 
Philip Kohn 
Pascal Koiran 
Adam Kowalczyk 
Anders Krogh 
Ramon Krosley 
Kevin Lang 
Steve Lawrence 
John Lazzaro 
M.D. Levine 
Tsungnan Lin 
Shawn Lockery 
David Lowe 
Stefan Manke 
Bimal Mathur 
Bartlett Mel 
Stefan Miesbach 
Kenneth Miller 
Manavendra Misra 
Melanie Mitchell 
Martin Moller 
F. Read Montague 
Andrew Moore 
Nelson Morgan 
J. Anthony Movshon 
Sayandev Mukherjee 
Ralph Neuneier 
Ernst Niebur 
Steven Nowlan 
Antonio Olmos-Gallo 
Christian Omlin 
Manfred Opper 
Barak Pearlmutter 
Fernando Pereira 
Jim Peterson 
Tony Plate 
John Platt 
David Plaut 
Jordan Pollack 
Dean Pomerleau 
Larry Reeker 
Steve Renals 
Barry Richmond 
Tony Robinson 
Dana Ron 
Eytan Ruppin 
Philip Sabes 
E. Sackinger 
Majd Sakr 
Lawrence Saul 
Eric Saund 
Terrence Sejnowski 
Sebastian Seung 
Robert Shapley 
Noel Sharkey 
Jude Shavlik 
Tadashi Shibata 
Patrice Simard 
Yoram Singer 
Massimo Sivilotti 
Warren Smith 
Padhraic Smyth 
Bill Softky 
David Standley 
Martin Stemmler 
Richard Sutton 
Sebastian Thmn 
Geoffrey Towell 
Ah Chung Tsoi 
Michael Turmon 
Lyle Ungar 
Santosh Venkatesh 
Grace Wahba 
DeLiang Wang 
Chris Watkins 
Raymond Watrous 
John Wawrzynek 
Ronald Williams 
Robert Williamson 
Charles Wilson 
David Wolpert 
Lei Xu 
Hezy Yeshurun 
ADVANCES IN NEURAL INFORMATION 
PROCESSING SYSTEMS 8 
PART I 
COGNITIVE SCIENCE 
