MeetingsFrom: Rich Caruana caruana+@cs.cmu.eduDate: Wed, 04 Oct 2000 12:25:55 -0400 Subject: Post-NIPS*2000 Workshops, Breckenridge, CO, December 1-2, 2000 The NIPS*2000 Workshops will be held Friday and Saturday, December 1 and 2, in Breckenridge Colorado after the NIPS conference in Denver Monday-Thursday, November 27-30. This year there are 18 workshops: - Affective Computing - Algorithms and Technologies for Neuroprosthetics and Neurorobotics - Computation in the Cortical Column - Computational Molecular Biology - Computational Neuropsychology - Cross-Validation, Bootstrap, and Model Selection - Data Fusion -- Theory and Applications - Data Mining and Learning on the Web - Explorative Analysis and Data Modeling in Functional Neuroimaging - Geometric Methods in Learning Theory - Information and Statistical Structure in Spike Trains - Learn the Policy or Learn the Value-Function? - New Perspectives in Kernel-based Learning Methods - Quantum Neural Computing - Representing the Structure of Visual Objects - Real-Time Modeling for Complex Learning Tasks - Software Support for Bayesian Analysis Systems - Using Unlabeled Data for Supervised Learning All workshops are open to all registered attendees. Many workshops also invite submissions. Submissions, and questions about individual workshops, should be directed to each workshop's organizers. Included below is a short description of each workshop. Additional information is available at the NIPS*2000 Workshop Web Page: http://www.cs.cmu.edu/Groups/NIPS/NIPS2000/Workshops/ Information about registration, travel, and accommodations for the main conference and the workshops is available at: http://www.cs.cmu.edu/Web/Groups/NIPS/ Breckenridge is a ski resort a few hours drive from Denver. The daily workshop schedule is designed to allow participants to ski half days, or enjoy other extra-curricular activities. Some may wish to extend their visit to take advantage of the relatively low pre-season rates. We look forward to seeing you in Breckenridge. Rich Caruana and Virginia de Sa NIPS Workshops Co-chairs * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Affective Computing Workshop Co-Chairs: Javier R. Movellan, Institute for Neural Computation, UCSD Marian Bartlett, Institute for Neural Computation, UCSD Gary Cottrell, Computer Science, UCSD Rosalind W. Picard, Media Lab, MIT Description: The goal of this workshop is to explore and discuss the idea of affective computers, i.e., computers that have the ability to express emotions, recognize emotions, and whose behavior is modulated by emotional states. Emotions are a fundamental part of humans intelligence. It may be argued that emotions provide an "operating system" for autonomous agents that need to handle uncertainty of natural environments in a flexible and efficient manner. Connectionist models of emotion dynamics have been developed (Velasquez, 1996) providing examples of computational systems that incorporate emotional dynamics and that are being used in actual autonomous agents (e.g., robotic pets). Emotional skills, especially the ability to recognize and express emotions, are essential for natural communication between humans, and until recently, have been absent from the computer side of the human-computer interaction. For example, autonomous teaching agents and pet robots would greatly benefit from detecting affective cues from the users (curiosity, frustration, insight, anger) and adjusting to them, and also from displaying emotion appropriate to the context. The workshop will bring together leaders in the main research areas of affective computing: emotion recognition, emotion synthesis, emotion dynamics, applications. Speakers from industry will discuss current applications of affective computing, including synthesizing facial expressions in the entertainment industry, increasing the appeal of the pet robots through emotion recognition and synthesis, and measuring galvanic skin response through the mouse to determine user frustration. Format: This will be a one day workshop. The speakers will be encouraged to talk about challenges and controversial topics both in their prepared talks and in the insuing discussions. Since one of the goals of the workshop is to facilitate communication between researchers in different subfields, ample time will be given to questions. The last part of the workshop will be devoted to a discussion of the most promising approaches and ideas that will have emerged during the workshop. Contact Info: Javier R. Movellan Institute for Neural Computation University of California San Diego La Jolla, CA 92093-0515 movellan@inc.ucsd.edu Marian Stewart Bartlett Institute for Neural Computation University of California San Diego La Jolla, CA 92093-0515 marni@inc.ucsd.edu * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Algorithms, technologies, and neural representations for neuroprosthetics and Neurorobotics. Organizers: Simon F Giszter and Karen A Moxon http://neurobio.mcphu.edu/GiszterWeb/nips_ws_2k.html Goals and objectives: The goal of the workshop is to bring together researchers interested in neuroprosthetics and neurorobotics and intermediate representations in the brain with a view to generating a lively discussion of the design principles for a brain to artificial device interface. Speakers will be charged to address (favourably or unfavourably) the idea that the nervous system is built around, or dynamically organizes, low dimensional representations which may be used in (or need to be designed into) neurosprosthetic interfaces and controllers. Some current prosthetics are built around explicit motor representations e.g. kinematic plans. Though controversial, the notion of primitives and low dimensional representations of input and output are gaining favor. These may or may not contain or be used in explicit plans. It is very likely that the appropriate choices of sensory and motor representations and motor elements are critical for the design of an integrated sensory motor prostheses that enables rapid adaptive learning, and creative construction of new motions and planning and execution. With a burgeoning interest in neuroprosthetics it is therefore timely to address how the interfaces to neuroprostheses should be conceptualized: what reprsentations should be extracted, what control elements should be provided and how should these be integrated. We hope to engage a wide range of perspectives to address needs for research and the possibilities enabled by neuroprosthetics. We intend to assemble presentations and discussions from the perspectives of both neural data and theory, of new technologies and algorithms, and of applications or experimental approaches enabled by new and current technologies. Anticipated or fully confirmed speaker/ participants: John Chapin: Neurorobotics Nikos Hatzopoulos: Neural coding in cortex of primates Scott Makeig or Terry Sejnowski: EEG based controllers: representations James Abbas: peripheral FES with CPG models Warren Grill and Michel Lemay Intraspinal FES and force-fields Karen Moxon : sensory prostheses Gerry Loeb : intramuscular prostheses and spinal controls Simon Giszter : spinal primitives and interface Emo Todorov : cortical encoding and representation Igo Krebs : rehabilitation with robots * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * COMPUTATION IN THE CORTICAL COLUMN Organizers: Henry Markram and Jennifer Linden URL: http://www.keck.ucsf.edu/~linden/ColumnWorkshop.html Understanding computation in the cortical column is a holy grail for both experimental and theoretical neuroscience. The basic six-layered neocortical columnar microcircuit, implemented most extensively (and perhaps in its most sophisticated form) in the human brain, supports a huge variety of sensory, cognitive, and motor functions. The secret behind the incredible flexibility and power of cortical columns has remained elusive, but new insights are emerging from several different areas of research. It is a great time for cortical anatomists, physiologists, modellers, and theoreticians to join forces in attempting to decipher computation in the cortical column. In this workshop, leading experimental and theoretical neuroscientists will present their own visions of computation in the cortical column, and will debate their views with an interdisciplinary audience. During the morning session, speakers and panel members will analyze columnar computation from their perspectives as authorities on the anatomy, physiology, evolution, and network properties of cortical microcircuitry. Speakers and panelists in the afternoon session will consider the functional significance of the cortical column in light of their expert knowledge of two columnar systems which have attracted intensive experimental attention to date: the visual cortex of cats and primates, and the barrel cortex of rodents. The goal of the workshop will be to define answers to four questions. ANATOMY: Does a common denominator, a repeating microcircuit element, exist in all neocortex? PHYSIOLOGY: What are the electrical dynamics, the computations, of the six-layered cortical microcircuit? FUNCTION: How do cortical columns contribute to perception? EVOLUTION: How does the neocortex confer such immense adaptability? * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Computational Molecular Biology Organizers: Tommi Jaakkola, MIT Nir Friedman, Hebrew University For more information contact the workshop organizers at: tommi@ai.mit.edu nir@cs.huji.ac.il * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * NIPS*2000 Workshop on Computational Neuropsychology Workshop Organizers: Sara Solla Northwestern University Michael Mozer University of Colorado Martha Farah University of Pennsylvania The 1980's saw two important developments in the sciences of the mind: The development of neural network models in cognitive psychology, and the rise of cognitive neuroscience. In the 1990's, these two separate approaches converged, and one of the results was a new field that we call "Computational Neuropsychology." In contrast to traditional cognitive neuropsychology, computational neuropsychology uses the concepts and methods of computational modeling to infer the normal cognitive architecture from the behavior of brain-damaged patients. In contrast to traditional neural network modeling in psychology, computational neuropsychology derives constraints on network architectures and dynamics from functional neuroanatomy and neurophysiology. Unfortunately, work in computational neuropsychology has had relatively little contact with the Neural Information Processing Systems (NIPS) community. Our workshop aims to expose the NIPS community to the unusual patient cases in neuropsychology and the sorts of inferences that can be drawn from these patients based on computational models, and to expose researchers in computational neuropsychology to some of the more sophisticated modeling techniques and concepts that have emerged from the NIPS community in recent years. We are interested in speakers from all aspects of neuropsychology, including: * attention (neglect) * visual and auditory perception (agnosia) * reading (acquired dyslexia) * face recognition (prosopagnosia) * memory (Alzheimer's, amnesia, category-specific deficits) * language (aphasia) * executive function (schizophrenia, frontal deficits). Further information about the workshop can be obtained at: http://www.cs.colorado.edu/~mozer/nips2000workshop.html Contact Sara Solla (solla@nwu.edu) or Mike Mozer (mozer@colorado.edu) if you are interested in speaking at the workshop. * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Call for Papers NIPS*2000 Workshop: Cross-Validation, Bootstrap and Model Selection Organizers: Rahul Sukthankar Compaq CRL and Robotics Institute, Carnegie Mellon Larry Wasserman Department of Statistics, Carnegie Mellon Rich Caruana Center for Automated Learning and Discovery, Carnegie Mellon Electronic Submission Deadline: October 18, 2000 (extended abstracts) Description Cross-validation and bootstrap are popular methods for estimating generalization error based on resampling a limited pool of data, and have become widely-used for model selection. The aim of this workshop is to bring together researchers from both matchine learning and statistics in an informal setting to discuss current issues in resampling-based techniques. These include: * Improving theoretical bounds on cross-validation, bootstrap or other resampling-based methods; * Empirical or theoretical comparisons between resampling-based methods and other forms of model selection; * Exploring the issue of overfitting in sampling-based methods; * Efficient algorithms for estimating generalization error; * Novel resampling-based approaches to model selection. The format for this one day workshop consists of invited talks, a panel discussion and short presentations from accepted submissions. Participants are encouraged to submit extended abstracts describing their current research in this area. Results presented at other conferences are eligible, provided that they are of broad interest to the community and clearly identified as such. Submissions for workshop presentations must be received by October 18, 2000, and should be sent to rahuls=nips@cs.cmu.edu. Extended abstracts should be in Postscript or Acrobat format and 1-2 pages in length. Contact Information The workshop organizers can be contacted by email at rahuls=nips@cs.cmu.edu, or at the phone/fax numbers listed below. Organizer Email Phone Fax Rahul Sukthankar rahuls@cs.cmu.edu +1-617-551-7694 +1-617-551-7650 Larry Wasserman larry@stat.cmu.edu +1-412-268-8727 +1-412-268-7828 Rich Caruana caruana@cs.cmu.edu +1-412-268-7664 * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Workshop Title: Data Fusion -- Theory and Applications Multisensor data fusion refers to the acquisition, processing, and synergistic combination of information gathered by various knowledge sources and sensors to provide a better understanding of the phenomenon under consideration. The concept of fusion underlies many information processing mechanisms in machines and in biological systems. In biological/perceptual systems, information fusion seems to account for remarkable performance and robustness when confronted with a variety of uncertainties. The complexity of fusion processes is due to many factors including uncertainties associated with different information sources, complimentarily of individual sources. For example, modeling, processing, fusion, and interpretation of diverse sensor data for knowledge assimilation and inferencing pose challenging problems, especially when available information is incomplete, inconsistent, and/or imprecise. The potential for significantly enhanced performance and robustness has motivated vigorous ongoing research in both biological and artificial multisensor data fusion algorithms, architectures, and applications. Such efforts deal with fundamental issues including modeling process, architecture and algorithms, information extraction, fusion process, optimization of fused performance, real time (dynamic) fusion etc. The goal of this workshop is to bring together researchers from various diverse fields (learning, human computer interaction, vision, speech, neural biology, etc) to discuss both theoretical and application issues that are relevant across different fields. It aims is to make the NIPS community aware of the various aspects and current status of this field, as well as the problems that remain unsolved. We are calling for participations. Submissions should be sent to the workshop organizers. Workshop Organizers: Misha Pavel pavel@ece.ogi.edu (503)748-1155 (o) Dept. of Electrical and Computer Engineering Oregon Graduate Institute of Science and Technology 20000 NW Walker Road Beaverton, OR 97006 Xubo Song xubosong@ece.ogi.edu (503) 748-1311 (o) Dept. of Electrical and Computer Engineering Oregon Graduate Institute of Science and Technology 20000 NW Walker Road Beaverton, OR 97006 Workshop Web Page: http://www.ece.ogi.edu/~xubosong/FusionWorkshop.html * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Data Mining and Learning on the Web Organizers: Gary William Flake (flake@research.nj.nec.com), Frans Coetzee (coetzee@research.nj.nec.com), and David Pennock (dpennock@research.nj.nec.com) No doubt about it, the web is big. So big, in fact, that many classical algorithms for databases and graphs cannot scale to the distributed multi-terabyte anarchy that is the web. How, then, do we best use, mine, and model this rich collection of data? Arguably, the best approach to developing scalable non-trivial applications and algorithms is to exploit the fact that, both as a graph and as a database, the web is highly non-random. Furthermore, since the web is mostly created and organized by humans, the graph structure (in the form of hyperlinks) encodes aspects of the content, and vice-versa. =20 We will discuss methods that exploit these properties of the web. We will further consider how many of the classical algorithms, which were formulated to minimize worst-case performance over all possible problem instances, can be adapted to the more regular structure of the web. Finally, we will attempt to identify major open research directions. This workshop will be organized into three mini-sessions: (1) Systematic Web Regularities (2) Web Mining Algorithms, and (3) Inferable Web Regularities. All speakers are invited and a partial list of confirmed speakers includes: Albert-L=E1szl=F3 Barab=E1si, Justin Boyan, Rich Caruana, Soumen Chakrabarti, Monika Henzinger, Ravi Kumar, Steve Lawrence, and Andrew McCallum. * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * EXPLORATIVE ANALYSIS AND DATA MODELING IN FUNCTIONAL NEUROIMAGING: Arriving at, not starting with a hypothesis Advanced analysis methods and models of neuroimaging data are only marginally represented at the big international brain mapping meetings. This contrasts the broad belief in the Neuroimaging community that these approaches are crucial in the further development of the field. The purpose of this NIPS workshop is to bring together theoreticians developing and applying new methods of neuroimaging data interpretation. The workshop focuses on explorative analysis (a) and modeling (b) of neuroimaging data: a) Higher-order explorative analysis (for example: ICA/PCA, clustering algorithms) can reveal data properties in a data-driven, not hypothesis-driven manner. b) Models for neuroimaging data can guide the data interpretation. The universe of possible functional hypotheses can be constrained by models linking the data to other empirical data, for instance anatomical connectivity (pathway analysis to yield effective connectivity), encephalograghy data, or to behavior (computational function). The talks will introduce the new approaches, in discussions we hope to address benefits and problems of the various methods and of their possible combination. It is intended to discuss not only the theory behind the various approaches, but as well their value for improved data interpretation. Therefore, we strongly encourage theparticipation of neuroimaging experimenters concerned with functional paradigms suggesting the use of nonstandard data interpretation methods. More information can be found on the workshop webpage: http://www.informatik.uni-ulm.de/ni/staff/FSommer/workshops/nips_ws00.html For any requests please contact the workshop organizers: Fritz Sommer and Andrzej Wichert Department of Neural Information Processing University of Ulm D-89069 Ulm Germany Tel. 49(731)502-4154 49(731)502-4257 FAX 49(731)502-4156 * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Geometric and Quantum Methods in Learning Organization: Shunichi Amari, Amir Assadi (Chair), and Tomaso Poggio Description. The purpose of this workshop is to attract the attention of the learning community to geometric methods and to take on an endeavor to: 1. lay out a geometric paradigm for formulating profound ideas in learning; 2. to facilitate the development of geometric methods suitable of investigation of new ideas in learning theory. Today's continuing advances in computation make it possible to infuse geometric ideas into learning that would otherwise have been computationally prohibitive. Quantum computation has created great excitement, offering a broad spectrum of new ideas for discovery of parallel-distributed algorithms, a hallmark of learning theory. In addition, geometry and quantum computation together offer a more profound picture of the physical world, and how it interacts with the brain, the ultimate learning system. Among the discussion topics, we envision the following: Information geometry, differential topological and quantum methods for turning local estimates into global quantities and invariants, Riemannian geometry and Feynman path integration as a framework to explore nonlinearity, and information theory of massive data sets. We will also examine the potential impact of learning theory on future development of geometry, and examples of how quantum computation has opened new vistas on design of parallel-distributed algorithms. The participants of the Workshop on Quantum Computation will find this workshop's geometric ideas beneficial for the theoretical aspects of quantum algorithms and quantum information theory. We plan to prepare a volume based on the materials for the workshops and other contributions to be proposed to the NIPS Program Committee. Contact Information Amir Assadi University of Wisconsin-Madison. URL: www.cms.wisc.edu/~cvg E-mail: ahassadi@facstaff.wisc.edu Partial List of Speakers and Panelists Shun-Ichi Amari Amir Assadi Zubin Ghahramani Geoffrey Hinton Tomaso Poggio Jose Principe Scott Mackeig Naoki Saito (tentative) * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Information and Statistical Structure in Spike Trains: How can we calculate what we really want to know? Organizer: Jonathan D. Victor jdvicto@med.cornell.edu Advances in understanding how neurons represent and manipulate information in their spike trains will require a combination of appropriate theoretical, computational, and experimental strategies. The workshop has several aims: (1) By presenting currently available methods in a tutorial-like fashion, we hope to lower the energy barrier to experimentalists who are interested in using information-theoretic and related approaches, but have not yet done so. The presentation of current methods is to be done in a manner that emphasizes the theoretical underpinnings of different strategies and the assumptions and tradeoffs that they make. (2) By provide a forum for open discussion among current practitioners, we hope to make progress towards understanding the relationships of the available techniques, guidelines for their application, and the basis of the differences in findings across preparations. (3) By presenting the (not fully satisfactory) state of the art to an audience that includes theorists, we hope to spur progress towards the development of better techniques, with a particular emphasis on exploiting more refined hypotheses for spike train structure, and developing techniques that are applicable to multi-unit recordings. A limited number of slots are available for contributed presentations. Individuals interested in presenting a talk (approximately 20 minutes, with 10 to 20 minutes for discussion) should submit a title and abstract, 200-300 words, to the organizer by October 22, 2000. Please indicate projection needs (overheads, 2x2 slides, LCD data projector). For further information, please see http://www-users.med.cornell.edu/~jdvicto/nips2000.html * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Title: ====== Reinforcement Learning: Learn the Policy or Learn the Value-Function? Organising Committee ==================== Peter Bartlett (Peter.Bartlett@anu.edu.au) Jonathan Baxter (jbaxter@whizbang.com) David McAllester (dmac@research.att.com) Home page ========= http://csl.anu.edu.au/~bartlett/rlworkshop Workshop Outline ============== There are essentially three main approaches to reinforcement learning in large state spaces: 1) Learn an approximate value function and use that to generate a policy, 2) Learn the parameters of the policy directly, typically using a Monte-Carlo estimate of the performance gradient, and 3) "Actor-Critic" methods that seek to combine the best features of 1) and 2). There has been a recent revival of interest in this area, with many new algorithms being proposed in the past two years. It seems the time is right to bring together researchers for an open discussion of the three different approaches. Submissions are sought on any topic of related interest, such as new algorithms for reinforcement learning, but we are particularly keen to solicit contributions that shed theoretical or experimental light on the relative merits of the three approaches, or that provide a synthesis or cross-fertilization between the different disciplines. Format ====== There will be invited talks and a series of short contributed talks (15 minutes), with plenty of discussion time. If you are interested in presenting at the workshop, please send a title and short abstract to jbaxter@whizbang.com * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * New Perspectives in Kernel-based Learning Methods Nello Cristianini, John Shawe-Taylor, Bob Williamson http://www.cs.rhbnc.ac.uk/colt/nips2000.html Abstract: The aim of the workshop is to present new perspectives and new directions in kernel methods for machine learning. Recent theoretical advances and experimental results have drawn considerable attention to the use of kernel functions in learning systems. Support Vector Machines, Gaussian Processes, kernel PCA, kernel Gram-Schmidt, Bayes Point Machines, Relevance and Leverage Vector Machines, are just some of the algorithms that make crucial use of kernels for problems of classification, regression, density estimation, novelty detection and clustering. At the same time as these algorithms have been under development, novel techniques specifically designed for kernel-based systems have resulted in methods for assessing generalisation, implementing model selection, and analysing performance. The choice of model may be simply determined by parameters of the kernel, as for example the width of a Gaussian kernel. More recently, however, methods for designing and combining kernels have created a toolkit of options for choosing a kernel in a particular application. These methods have extended the applicability of the techniques beyond the natural Euclidean spaces to more general discrete structures. The field is witnessing growth on a number of fronts, with the publication of books, editing of special issues, organization of special sessions and web-sites. Moreover, a convergence of ideas and concepts from different disciplines is occurring. The growth is concentrated in four main directions: 1) design of novel kernel-based algorithms 2) design of novel types of kernel functions 3) development of new learning theory concepts 4) application of the techniques to new problem areas Extended abstracts may be submitted before October 30th to nello@dcs.rhbnc.ac.uk * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Quantum Neural Computing http://web.physics.twsu.edu/behrman/NIPS.htm Recently there has been a resurgence of interest in quantum computers because of their potential for being very much smaller and faster than classical computers, and because of their ability in principle to do heretofore impossible calculations, such as factorization of large numbers in polynomial time. This workshop will explore ways to implement quantum computing in network topologies, thus exploiting both the intrinsic advantages of quantum computing and the adaptability of neural computing. Aspects/approaches to be explored will include: quantum hardware, e.g. nmr, quantumdots, and molecular computing; theoretical and practical limits to quantum and quantum neural computing, e.g., noise and measurability; and simulations. Targeted groups: computer scientists, physicists and mathematicians interested in quantum computing and next-generation computing hardware. Invited speakers will include: Paul Werbos, NSF Program Director, Control, Networks & Computational Intelligence Program, Electrical and Communications Systems Division, who will keynote the workshop. Thaddeus Ladd, Stanford, "Crystal lattice quantum computation." Mitja Perus, Institute BION, Stegne 21, SI-1000 Ljubljana, Slovenia, "Quantum associative nets: A new phase processing model" Ron Spencer, Texas A&M University, "Spectral associative memories." E.C. Behrman, J.E. Steck, and S.R. Skinner, Wichita State University, "Simulations of quantum neural networks." Dan Ventura, Penn State: "Linear optics implementation of quantum algorithms." Ron Chrisley, TBA Send contributed papers, by October 20th, to: Co-chairs: Elizabeth C. Behrman behrman@wsuhub.uc.twsu.edu James E. Steck steck@bravo.engr.twsu.edu This Workshop is partially supported by the National Science Foundation, Grant #ECS-9820606. * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Workshop title: Representing the Structure of Visual Objects Web page: http://kybele.psych.cornell.edu/~edelman/NIPS00/index.html Organizers: Nathan Intrator (Nathan_Intrator@brown.edu) Shimon Edelman (se37@cornell.edu) Confirmed invited speakers: Ron Chrisley (Sussex) John Hummel (UCLA) Christoph von der Malsburg (USC) Pietro Perona (Caltech) Tomaso Poggio (MIT) Greg Rainer (Tuebingen) Manabu Tanifuji (RIKEN) Shimon Ullman (Weizmann) Description: The focus of theoretical discussion in visual object processing has recently started to shift from problems of recognition and categorization to the representation of object structure. The main challenges there are productivity and systematicity, two traits commonly attributed to human cognition. Intuitively, a cognitive system is productive if it is open-ended, that is, if the set of entities with which it can deal is, at least potentially, infinite. Systematicity, even more than productivity, is at the crux of the debate focusing on the representational theory of mind. A visual representation could be considered systematic if a well-defined change in the spatial configuration of the object (e.g., swapping top and bottom parts) were to cause a principled change in the representation (the representations of top and bottom parts are swapped). In vision, this issue (as well as compositionality, commonly seen as the perfect means of attaining systematicity) is, at present, wide open. The workshop will start with an introductory survey of the notions of productivity, systematicity and compositionality, and will consist of presentations by the proponents of some of the leading theories in the field of structure representation, interspersed with open-floor discussion. * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Title: Real-Time Modeling for Complex Learning Tasks The goal of this workshop is to develop a better understanding how to create new statistical learning techniques that can deal with complex, high dimensional data sets, where (possibly redundant and/or irrelevant) data is received continuously from sensors and needs to be incorporated in learning models that may have to change their structure during learning under real time constraints. The workshop aims at bringing together researchers from various theoretical learning frameworks (Bayesian, Nonparametric statistics, Kernel methods, Gaussian processes etc.) and application domains to discuss future research direction for principled approaches towards real-time learning. For further details, please refer to the URL: http://www-slab.usc.edu/events/NIPS2000 * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Software Support for Bayesian Analysis System URL: http://ase.arc.nasa.gov/nips2000 Bayesian Analysis is an established technique for many data analysis applications. The development of application software for any specific analysis problem, however, is a difficult and time-consuming task. Programs must be tailored to the specific problem, need to represent the given statistical model correctly, and should preferably run efficiently. Over the last years, a variety of different libraries, shells, and synthesis systems for Bayesian data analysis has been implemented which are intended to simplify application software development. The goal of this workshop is to bring together developers of such generic Bayesian software packages and tools (e.g., JavaBayes, AutoClass, BayesPack, BUGS, BayesNet Toolbox, PDP++) together with the developers of generic algorithm schemas (more recent ones amenable to automated effort include Structural EM, Fisher Kernel method, mean-field, etc.), and software engineering experts. It is intended as a forum to discuss and exchange the different technical approaches as for example usage of libraries, interpretation of statistical models (e.g., Gibbs sampling), or software synthesis based on generic algorithm schemas. The workshop aims to discuss the potential and problems of generic tools for the development of efficient Bayesian data analysis software tailored towards specific applications. If you are planning to attend this workshop as a particpiant and/or are interested to present your work, please send a short (1-4 pages) system description, technical paper, or position paper to fisch@ptolemy.arc.nasa.gov no later than Wednesday, October, 18 2000. Preliminary PC: Organizers: L. Getoor, Stanford W. Buntine, Dynaptics P. Smyth, UC Irvine B. Fischer, RIACS/NASA Ames M. Turmon, JPL J. Schumann, RIACS/NASA Ames K. Murphy, UC Berkeley * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * NIPS 2000 Workshop and Unlabeled Data Supervised Learning Competition! We are pleased to announce the NIPS 2000 Unlabeled Data Supervised Learning Competition! This competition is designed to compare algorithms and architectures that use unlabeled data to help supervised learning, and will culminate in a NIPS workshop, where approaches and results will be compared. Round three begins soon, so don't delay (it is also still possible to submit results for rounds 1 and 2). More details, are now available at the competition web-site: http://q.cis.uoguelph.ca/~skremer/NIPS2000/ May the best algorithm win! Stefan, Deb, and Kristin * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * |
Copyright © 2000 KDnuggets. Subscribe to KDnuggets News!