neuron-request@HPLABS.HP.COM ("Neuron-Digest Moderator Peter Marvit") (06/02/89)
Neuron Digest Friday, 2 Jun 1989 Volume 5 : Issue 25 Today's Topics: Pinker and Prince Bruce McNaughton on Neural Net for Spatial Rep. in Hippocampus TR - Choosing Computational Architectures for Text Processing Call for paper -- hybrid systems Workshop on Neural Representation of Visual Information Tech reports available conference announcement - EMCSR 1990 TD Model of Conditioning -- Paper Announcement Updated program info for: NEURAL NETWORKS for DEFENSE Watrous to speak at GTE Labs Send submissions, questions, address maintenance and requests for old issues to "neuron-request@hplabs.hp.com" or "{any backbone,uunet}!hplabs!neuron-request" ARPANET users can get old issues via ftp from hplpm.hpl.hp.com (15.255.16.205). ------------------------------------------------------------ Subject: Pinker and Prince From: marchman@amos.ling.ucsd.edu (Virginia Marchman) Date: Wed, 26 Apr 89 05:11:31 +0000 I heard that there was a conversation going on about the Pinker & Prince article, and thought that I would pass along an abstract from a recent Tech Report. Requests for hard copy should be sent to yvonne@amos.ucsd.edu. (ask for TR #8902). -virginia marchman Pattern Association in a Back Propagation Network: Implications for Child Language Acquisition Kim Plunkett Virginia Marchman University of Aarhus, Denmark University of California, San Diego Abstract A 3-layer back propagation network is used to implement a pattern association task which learns mappings that are analogous to the present and past tense forms of English verbs, i.e., arbitrary, identity, vowel change, and suffixation mappings. The degree of correspondence between connectionist models of tasks of this type (Rumelhart & McClelland, 1986; 1987) and children's acquisition of inflectional morphology has recently been highlighted in discussions of the general applicability of PDP to the study of human cognition and language (Pinker & Mehler, 1988). In this paper, we attempt to eliminate many of the shortcomings of the R&M work and adopt an empirical, comparative approach to the analysis of learning (i.e., hit rate and error type) in these networks. In all of our simulations, the network is given a constant 'diet' of input stems -- that is, discontinuities are not introduced into the learning set at any point. Four sets of simulations are described in which input conditions (class size and token frequency) and the presence/absence of phonological subregularities are manipulated. First, baseline simulations chart the initial computational constraints of the system and reveal complex "competition effects" when the four verb classes must be learned simultaneously. Next, we explore the nature of these competitions given different type (class sizes) and token frequencies (# of repetitions). Several hypotheses about input to children are tested, from dictionary counts and production corpora. Results suggest that relative class size determines which "default" transformation is employed by the network, as well as the frequency of overgeneralization errors (both "pure" and "blended" overgeneralizations). A third series of simulations manipulates token frequency within a constant class size, searching for the set of token frequencies which results in "adult-like competence" and "child-like" errors across learning. A final series investigates the addition of phonological sub-regularities into the identity and vowel change classes. Phonological cues are clearly exploited by the system, leading to overall improved performance. However, overgeneralizations, U-shaped learning and competition effects continue to be observed in similar conditions. These models establish that input configuration plays a role in detemining the types of errors produced by the network - including the conditions under which "rule-like" behavior and "U-shaped" development will and will not emerge. The results are discussed with reference to behavioral data on children's acquisition of the past tense and the validity of drawing conclusions about the acquisition of language from models of this sort. ------------------------------ Subject: Bruce McNaughton on Neural Net for Spatial Rep. in Hippocampus From: Mark Gluck <netlist@PSYCH.STANFORD.EDU> Date: Thu, 27 Apr 89 08:50:52 -0700 [[ Editor's Note: Although this talk has past, I continue the practice of including talks so readers (from all over the world) know what's going on and who is doing it. -PM ]] Stanford University Interdisciplinary Colloquium Series: Adaptive Networks and their Applications May 2nd (Tuesday, 3:30pm): Room 380-380C ****************************************************************************** Hebb-Steinbuch-Marr Networks and the Role of Movement in Hippocampal Representations of Spatial Relations Bruce L. McNaughton Dept. of Psychology University of Colorado Campus Box 345 Boulder, CO 80309 ****************************************************************************** Abstract Over 15 years ago, Marr proposed models for associative learning and pattern completion in specific brain regions. These models incorporated Hebb's postulate, the "learning matrix" concept of Steinbuch, recurrent excitation, and the assumptions that a few excitatory synapses are disproportionately powerful, and that inhibitory synapses divide postsynaptic excitation by the total input. These ideas provide a basis for understanding much of the circuitry and physiology of the hippocampus, and will be used to suggest how spatial relationships are coded there by forming conditional associations between location and movement representations originating in the inferotemporal and parietal cortical systems respectively. References: - ----------- McNaughton, B. L. & Morris R.G.M. (1988). Hippocampal synaptic enhancement and information storage within a distributed memory system. Trends in Neurosci. 10:408-415. McNaughton, B. L. & Nadel, L. (in press, 1989). Hebb-Marr networks and the neurobiological representation of action in space. To appear in M. Gluck & D. Rumelhart (Eds.), Neuroscience and Connectionist Models, Erlbaum: Hillsdale, NJ Additional Information: - ---------------------- Location: Room 380-380C, which can be reached through the lower level between the Psychology and Mathematical Sciences buildings. Level: Technically oriented for persons working in related areas. Mailing lists: To be added to the network mailing list, netmail to netlist@psych.stanford.edu with "addme" as your subject header. For additional information, contact Mark Gluck (gluck@psych.stanford.edu). ------------------------------ Subject: TR - Choosing Computational Architectures for Text Processing From: Mike Oaksford <mike%epistemi.edinburgh.ac.uk@NSFnet-Relay.AC.UK> Date: Fri, 05 May 89 09:34:17 +0100 Note, that on this advertisment we cleaned up our communicative act, thanks for being so tolerant of our strange driving habits. Choosing Computational Architectures for Text Processing Keith Stenning and Mike Oaksford Centre for Cognitive Science, University of Edinburgh Tech Report EUCCS/RP-28 (A shorter version to appear in "Connectionist Approaches to Language", Reilly, R., and Sharkey, N. (eds) In this paper we investigate various criteria which bear on the choice of computational architectures for text processing. The principle role of the computational or cognitive architecture is to provide mechanisms for inference. In the study of text processing two forms of inference are fundamental, (i) the implicit elaborative inferences required for interpretation and (ii) explicit inferences which can be the subject of a text. We suggest that the decision of what architecture to employ in accounting for these inferential modes can not be made *a priori*. We argue that classical cognitive architectures based on logic and proof theory although eminently suited to (ii) fail to provide tractable theories of (i), while more recent proposals like PDP (Rumelhart & McClelland, 1986) and Classifier systems (Holland, Holyoak, Nisbett & Thagard, 1986), seem to offer new insights into (i) while leaving (ii) untouched. We examine the computational issues involved in a review of recent candidate architectures beginning at one extreme with PROLOG, going through ACT* and Classifier systems, ending with PDP. We then examine the empirical work from verbal reasoning tasks involving conditional and syllogistic reasoning in arguing that the grounds upon which to choose between architectures are largely *a posteriori* and empirical and moreover indicate that satisfactory explanations of this data must invoke both (i) and (ii). In the process we shall proffer novel interpretations both of conditional reasoning experiments as being largely inductive (and hence of scant relevance to assessing our facility for logical thought) and of Johnson-Laird's theory of syllogisms as providing a heuristic theorem prover along the lines of any other practical *implementation* of logic. We believe that this allows the explanatory burden for a lot of this data to be correctly located at the *implemenational* rather than the cognitive level. Requests for this Tech Report to: Betty Hughs, Technical Report Librarian, Centre for Cognitive Science, University of Edinburgh, 1 & 2, Buccleuch Place, Edinburgh, EH8 9LW, Scotland, UK. e-mail: betty%epistemi.ed.ac.uk@nsfnet-relay.ac.uk ------------------------------ Subject: Call for paper -- hybrid systems From: hendler@icsi.berkeley.edu (James Hendler) Organization: International Computer Science Institute Date: Tue, 09 May 89 20:34:48 +0000 CALL FOR PAPERS CONNECTION SCIENCE (Journal of Neural Computing, Artificial Intelligence and Cognitive Research) Special Issue -- HYBRID SYMBOLIC/CONNECTIONIST SYSTEMS Connectionism has recently seen a major resurgence of interest among both artificial intelligence and cognitive science researchers. The spectrum of connectionist approaches is quite large, ranging from structured models, in which individual network units carry meaning, through distributed models of weighted networks with learning algorithms. Very encouraging results, particularly in ``low-level'' perceptual and signal processing tasks, are being reported across the entire spectrum of these models. Unfortunately, connectionist systems have had more limited success in those ``higher cognitive'' areas where symbolic models have traditionally shown promise: expert reasoning, planning, and natural language processing. While it may not be inherently impossible for purely connectionist approaches to handle complex reasoning tasks someday, it will require significant breakthroughs for this to happen. Similarly, getting purely symbolic systems to handle the types of perceptual reasoning that connectionist networks perform well would require major advances in AI. One approach to the integration of connectionist and symbolic techniques is the development of hybrid reasoning systems in which differing components can communicate in the solving of problems. This special issue of the journal Connection Science will focus on the state of the art in the development of such hybrid reasoners. Papers are solicited which focus on: Current artificial intelligence systems which use connectionist components in the reasoning tasks they perform. Theoretical or experimental results showing how symbolic computations can be implemented in, or augmented by, connectionist components. Cognitive studies which discuss the relationship between functional models of higher level cognition and the ``lower level'' implementations in the brain. The special issue will give special consideration to papers sharing the primary emphases of the Connection Science Journal which include: 1) Replicability of Results: results of simulation models should be reported in such a way that they are repeatable by any competent scientist in another laboratory. The journal will be sympathetic to the problems that replicability poses for large complex artificial intelligence programs. 2) Interdisciplinary research: the journal is by nature multidisciplinary and will accept articles from a variety of disciplines such as psychology, cognitive science, computer science, language and linguistics, artificial intelligence, biology, neuroscience, physics, engineering and philosophy. It will particularly welcome papers which deal with issues from two or more subject areas (e.g. vision and language). Papers submitted to the special issue will also be considered for publication in later editions of the journal. All papers will be refereed. The expected publication date for the special issue is Volume 2(1), March, 1990. DEADLINES: Submission of papers June 15, 1989 Reviews/decisions September 30, 1989 Final rewrites due December 15, 1989. Authors should send four copies of the article to: Prof. James A. Hendler Associate Editor, Connection Science Dept. of Computer Science University of Maryland College Park, MD 20742 USA Those interested in submitting articles are welcome to contact the editor via e-mail (hendler@brillig.umd.edu - US Arpa or CSnet) or in writing at the above address. ------------------------------ Subject: Workshop on Neural Representation of Visual Information From: rapaport@CS.BUFFALO.EDU (William J. Rapaport) Organization: The Internet Date: Tue, 16 May 89 15:25:36 +0000 STATE UNIVERSITY OF NEW YORK AT BUFFALO UB VISION GROUP and GRADUATE RESEARCH INITIATIVE IN COGNITIVE AND LINGUISTIC SCIENCES invite you to attend a workshop: NEURAL REPRESENTATION OF VISUAL INFORMATION June 9, 8:30 am to 10 pm June 10, 8:30 am to 4 pm Lipschitz Room, CFS 126, Main Street Campus Speakers: Dana Ballard, Computer Science, Rochester Robet Boynton, Psychology, UC San Diego Ennio Mingola, Center for Adaptive Systems, Boston U. Ken Naka, National Inst. for Basic Biology, Japan, and NYU Hiroka Sakai, National Inst. for Basic Biology, Japan, and NYU Members of the UB Vision Group If you are interested in attending, send your name and address with a check for $40 to cover the cost of the five meals to: Dr. Deborah Walters Department of Computer Science SUNY Buffalo Buffalo, NY 14260 Graduate students may apply for a waiver of the meal fee. For further information, contact Dr. Walters, 636-3187, email: walters@cs.buffalo.edu or walters@sunybcs.bitnet. ------------------------------ Subject: Tech reports available From: GINDI%GINDI@Venus.YCC.Yale.Edu Date: Fri, 19 May 89 09:54:00 -0400 The following two tech reports are now available. Please send requests to GINDI@VENUS.YCC.YALE.EDU or by physical mail to: Gene Gindi Yale University Department of Electrical Engineering P.O. Box 2157 , Yale Station New Haven, CT 06520 ______________________________________________________________________ Yale University, Dept. Electrical Engineering Center for Systems Science TR- 8903 Neural Networks for Object Recognition within Compositional Hierarches: Initial Experiments Joachim Utans, Gene Gindi * Dept. Electrical Engineering Yale University P.O. Box 2157, Yale Station New Haven CT 06520 *(to whom correspondence should be addressed) Eric Mjolsness, P. Anandan Dept. Computer Science Yale University New Haven CT 06520 Abstract We describe experiments with TLville, a neural-network for object recognition. The task is to recognize, in a translation-invariant manner, simple stick figures. We formulate the recognition task as the problem of matching a graph of model nodes to a graph of data nodes. Model nodes are simply user-specified labels for objects such as "vertical stick" or "t-junction"; data nodes are parameter vectors, such as (x,y,theta), of entities in the data. We use an optimization approach where an appropriate objective function specifies both the graph-matching problem and an analog neural net to carry out the optimization. Since the graph structure of the data is not known a priori; it must be computed dynamically as part of the optimization. The match metrics are model-specific and are invoked selectively, as part of the optimization, as various candidate matches of model-to-data occur. The network supports notions of abstraction in that the model nodes express compositional hierarchies involving object-part relationships. Also, a data node matched to an whole object contains a dynamically computed parameter vector which is an abstraction summarizing the parameters of data nodes matched to the constituent parts of the whole. Terms in the match metric specify the desired abstraction. In addition, a solution to the problem of computing a transformation from retinal to object-centered coordinates to support recognition is offered by this kind of network; the transformation is contained as part of the objective function in the form of the match metric. In experiments, the network usually succeeds in recognizing single or multiple instances of a single composite model amid instances of non-models, but it gets trapped in unfavorable local minima of the 5th-order objective when multiple composite objects are encoded in the database. ______________________________________________________________________ Yale University, Dept. Electrical Engineering Center for Systems Science TR- 8908 Stickville: A Neural Net for Object Recognition via Graph Matching Grant Shumaker School of Medicine, Yale University, New Haven, CT 06510 Gene Gindi Department of Electrical Engineering, Yale University P.O. Box 2157 Yale Station, New Haven,CT 06520 (to whom correspondence should be addressed) Eric Mjolsness, P.Anandan Department of Computer Science, Yale University, New Haven, CT 06510 Abstract An objective function for model-based object recognition is formulated and used to specify a neural network whose dynamics carry out the optimization, and hence the recognition task. Models are specified as graphs that capture structural properties of shapes to be recognized. In addition, compositional (INA) and specialization (ISA) hierarchies are imposed on the models as an aid to indexing and are represented in the objective function as sparse matrices. Data are also represented as a graph. The optimization is a graph-matching procedure whose dynamical variables are ``neurons'' hypothesizing matches between data and model nodes. The dynamics are specified as a third-order Hopfield-style network augmented by hard constraints implemented by ``Lagrange multiplier'' neurons. Experimental results are shown for recognition in Stickville, a domain of 2-D stick figures. For small databases, the network successfully recognizes both an object and its specialization. ------------------------------ Subject: conference announcement - EMCSR 1990 From: mcvax!ai-vie!georg@uunet.UU.NET (Georg Dorffner) Date: Fri, 19 May 89 14:35:03 -0100 Announcement and Call for Papers EMCSR 90 TENTH EUROPEAN MEETING ON CYBERNETICS AND SYSTEMS RESEARCH April 17-20, 1990 University of Vienna, Austria Session M: Parallel Distributed Processing in Man and Machine Chairs: D.Touretzky (Carnegie Mellon, Pittsburgh, PA) G.Dorffner (Vienna, Austria) Other Sessions at the meeting will be: A: General Systems Methodology B: Fuzzy Sets, Approximate Reasoning and Knowledge-based Systems C: Designing and Systems D: Humanity, Architecture and Conceptualization E: Cybernetics in Biology and Medicine F: Cybernetics in Socio-Economic Systems G: Workshop: Managing Change: Institutional Transition in the Private and Public Sector H: Innovation Systems in Management and Public Policy I: Systems Engineering and Artificial Intelligence for Peace Research J: Communication and Computers K: Software Development for Systems Theory L: Artificial Intelligence N: Impacts of Artificial Intelligence The conference is organized by the Austrian Society for Cybernetic Studies (chair: Robert Trappl). SUBMISSION OF PAPERS: For symposium M, all contributions in the fields of PDP, connectionism, and neural networks are welcome. Acceptance of contributors will be determined on the basis of Draft Final Papers. These papers must not exceed 7 single-spaced A4 pages (maximum 50 lines, final size will be 8.5 x 6 inches), in English. They have to contain the final text to be submitted, however, graphs and pictures need not be of reproducible quality. The Draft Final Paper must carry the title, author(s) name(s), and affiliation in this order. Please specify the symposium in which you would like to present the paper (one of the letters above). Each scientist shall submit only 1 paper. Please send t h r e e copies of the Draft Final Paper to: EMCSR 90 - Conference Secretariat Austrian Society for Cybernetic Studies Schottengasse 3 A-1010 Vienna, Austria Deadline for submission: Oct 15, 1989 Authors will be notified about acceptance no later than Nov 20, 1989. They will then be provided with the detailed instructions for the preperation of the Final Paper. Proceedings containing all accepted papers will be printed. For further information write to the above address, call +43 222 535 32 810, or send email to: sec@ai-vie.uucp Questions concerning symposium M (Parallel Distributed Processing) can be directed to Georg Dorffner (same address as secretariat), email: georg@ai-vie.uucp ------------------------------ Subject: TD Model of Conditioning -- Paper Announcement From: Rich Sutton <rich@gte.com> Date: Fri, 19 May 89 15:01:15 -0400 Andy Barto and I have just completed a major new paper relating temporal-difference learning, as used, for example, in our pole-balancing learning controller, to classical conditioning in animals. The paper will appear in the forthcoming book ``Learning and Computational Neuroscience,'' edited by J.W. Moore and M. Gabriel, MIT Press. A preprint can be obtained by emailing to rich%gte.com@relay.cs.net with your physical-mail address. The paper has no abstract, but begins as follows: TIME-DERIVATIVE MODELS OF PAVLOVIAN REINFORCEMENT Richard S. Sutton GTE Laboratories Incorporated Andrew G. Barto University of Massachusetts This chapter presents a model of classical conditioning called the temporal-difference (TD) model. The TD model was originally developed as a neuron-like unit for use in adaptive networks (Sutton & Barto, 1987; Sutton, 1984; Barto, Sutton & Anderson, 1983). In this paper, however, we analyze it from the point of view of animal learning theory. Our intended audience is both animal learning researchers interested in computational theories of behavior and machine learning researchers interested in how their learning algorithms relate to, and may be constrained by, animal learning studies. We focus on what we see as the primary theoretical contribution to animal learning theory of the TD and related models: the hypothesis that reinforcement in classical conditioning is the time derivative of a composite association combining innate (US) and acquired (CS) associations. We call models based on some variant of this hypothesis ``time-derivative models'', examples of which are the models by Klopf (1988), Sutton & Barto (1981a), Moore et al (1986), Hawkins & Kandel (1984), Gelperin, Hopfield & Tank (1985), Tesauro (1987), and Kosko (1986); we examine several of these models in relation to the TD model. We also briefly explore relationships with animal learning theories of reinforcement, including Mowrer's drive-induction theory (Mowrer, 1960) and the Rescorla-Wagner model (Rescorla & Wagner, 1972). In this paper, we systematically analyze the inter-stimulus interval (ISI) dependency of time-derivative models, using realistic stimulus durations and both forward and backward CS--US intervals. The models' behaviors are compared with the empirical data for rabbit eyeblink (nictitating membrane) conditioning. We find that our earlier time-derivative model (Sutton & Barto, 1981a) has significant problems reproducing features of these data, and we briefly explore partial solutions in subsequent time-derivative models proposed by Moore et al. (1986), Klopf (1988), and Gelperin et al. (1985). The TD model was designed to eliminate these problems by relying on a slightly more complex time-derivative theory of reinforcement. In this paper, we motivate and explain this theory from the point of view of animal learning theory, and show that the TD model solves the ISI problems and other problems with simpler time-derivative models. Finally, we demonstrate the TD model's behavior in a range of conditioning paradigms including conditioned inhibition, primacy effects (Egger & Miller, 1962), facilitation of remote associations, and second-order conditioning. ------------------------------ Subject: Updated program info for: NEURAL NETWORKS for DEFENSE From: marvit@hplabs.hp.com Date: Wed, 24 May 89 08:51:26 -0700 [[ Editor's Note: Again, the originator's asked me to forward this message. Note Citizenship requirements. Please contact the folks listed at the end for further info. -PM ]] UPDATED PROGRAM INFORMATION FOR: --------------------------- NEURAL NETWORKS for DEFENSE A One-day Conference: --------------------------- Saturday, June 17, 1989 (the day before IJCNN) Washington, DC Conference Chair: Prof. Bernard Widrow (Stanford Univ.) ------------------------------------------------------- A one-day conference on defense needs, applications, and opportunities for computing with neural networks, featuring key representatives from government and industry. It will take place in Washington, DC, right before the IEEE and INNS's International Joint Conference on Neural Networks (IJCNN). INDUSTRY SESSION: The industry session will feature presentations of the current status of defense-oriented research, development, and applications of neural network technology from industry leaders. They will discuss their current, past, and future involvement in neural networks and defense technology, as well as the kinds of cooperative ventures in which they might be interested. * Patrick Castelaz. HUGHES AEROSPACE, Command and Control: "Signal Processing Applications of Neural Networks" * Richard Elsley. ROCKWELL INTERNATIONAL, Knowledge Systems Division: "Neural Network Research at Rockwell" * Lawrence Seidman. FORD AEROSPACE, Advanced Technology: "Overview of Neural Network Applications at Ford Aerospace" * James Anderson. BROWN UNIV., Dept. of Cognitive & Linguistic Sciences: "Adventures with TEXAS INSTRUMENTS and the U.S. Air Force in Radar Trasnmitter Categorization and Identification" * Robert Dawes. MARTINGALE RESEARCH: "Adaptive Bayesian Estimation and Control with Neural Networks" * Harold Stoll. NORTHROP, Integrated Optics Laboratory: "Optical Neural Networks for Automatic Target Recognition" * Michael Buffa. NESTOR CORP.: "Military Target Recognition in Sonar, Radar, and Image-Based Systems with Actual Results" * Fred Weingard. BOOZ, ALLEN, & HAMILTON, Neural Network Applications Group "The Adaptive Network Sensor Procesor Program at Wright-Patterson Air Force Base" * Patrick Simpson. GENERAL DYNAMICS: "Defense Applications of Neural Networks" * John Dishon. SAIC, Emerging Technologies Division "Non-Linear Adaptive Control Systems in Neural Networks" * Monndy Eshera. MARTIN MARIETTA: "Systolic Array Neurocomputers: Synaptic Level Parallelism" * Robert Hecht-Nielsen. HNC: "Near Term Defense Payoffs with Neurocomputing Technology" * John Leonard. HUGHES AEROSPACE, Electro-Optical & Data Systems Group: "Neural Networks for Tactical Awareness & Target Recognition" * Robert Willstadter. BOEING, Computer Services - AI Systems: "Defense & Aerospace Applications of Neural Networks at Boeing" DEFENSE DEPARTMENT SESSION: The defense-department session will include program managers from Department of Defense (DoD) agencies funding and conducting Neural Network research and development: * Thomas Mckenna. Scientific Officer, Cognitive & Neural Sciences: OFFICE OF NAVAL RESEARCH (ONR) * David Andes, Chief of AI & Neural Network Research Programs: NAVAL WEAPONS CENTER, CHINA LAKE: * Edward Gliatti, Chief of Information Processing Technology Branch WRIGHT-PATTERSON AIR FORCE BASE: * Major Robert L. Russel, Jr. Assistant Division Chief for Image Systems ROME AIR DEVELOPMENT CENTER: ...plus others to be announced later. KEY NOTE ADDRESS: The meeting chairmain, and the keynote speaker at the conference, is Professor Bernard Widrow who directed the recent DARPA study evaluating the military and commercial potential of neural networks. He is a professor of EE at Stanford University, the current president of the INNS, co-inventor of the LMS algorithm (Widrow & Hoff, 1960), and the president of Memistor Corp, the oldest neural network applications and development company, which Prof. Widrow founded in 1962. WHO SHOULD COME: High-technology research & development personnel, research directors, and management from defense-oriented R&D companies and divisions. Defense Department personnel in involved with, or directing, Neural Network research in areas of possible Neural Network applications, including: Automatic Target Recognition; Speach, Sonar, & Radar Classification, and Real-time Sensorimotor Control for Autonomous Robotic Applications. Neural Network researchers who wish to be aware of current and future defense needs and opportunities for technology transfer to applications. * * * Program Committee: Mark Gluck (Stanford) & Ed Rosenfeld Note: Attendance at "N. N. for Defense" is limited to U.S. Citizens Only ---------------------------------------------------------------------------- For registration and information, call Anastasia Mills at (415) 995-2471 or FAX: (415) 543-0256, or write to: Neural Network Seminars, Neural Network Seminars, Miller-Freeman, 500 Howard St., San Fran., CA 94105 ---------------------------------------------------------------------------- ------------------------------ Subject: Watrous to speak at GTE Labs From: rich@GTE.COM (Rich Sutton) Organization: GTE Laboratories, Waltham, MA Date: Thu, 25 May 89 14:57:28 +0000 Seminar Announcement PHONEME DISCRIMINATION USING CONNECTIONIST NETWORKS R. Watrous Dept. of Computer Science, Univ. of Toronto Siemens Research and Technology Laboratories The application of connectionist networks to speech recognition is assessed using a set of representative phonetic discrimination problems chosen with respect to a theory of phonetics. A connectionist network model called the Temporal Flow Model is defined which represents temporal relationships using delay links and permits general patterns of connectivity. It is argued that the model has properties appropriate for time varying signals such as speech. Networks are trained using gradient descent methods of iterative nonlinear optimization to reduce the mean squared error between the actual and the desired response of the output units. Separate network solutions are demonstrated for all eight phonetic discrimination problems for one male speaker. The network solutions are analyzed carefully and are shown in every case to make use of known acoustic phonetic cues. The network solutions vary in the degree to which they make use of context dependent cues to achieve phoneme recognition. The network solutions were tested on data not used for training and achieved an average accuracy of 99.5%. It is concluded that acoustic phonetic speech recognition can be accomplished using connectionist networks. - --------------------------------------------------------------------------- The talk will be at 11am on May 31 in the GTE Labs Auditorium. For further information contact Rich Sutton (Rich%gte.com@relay.cs.net or 617-466-4133). Non-GTE people should arrive early to be escorted to the auditoriumm. ------------------------------ End of Neurons Digest *********************