/ Events

Traditional pattern recognition techniques are intimately linked to the notion of "feature spaces." Adopting this view, each object is described in terms of a vector of numerical attributes and is therefore mapped to a point in a Euclidean (geometric) vector space so that the distances between the points reflect the observed (dis)similarities between the respective objects. This kind of representation is attractive because geometric spaces offer powerful analytical as well as computational tools that are simply not available in other representations. Indeed, classical pattern recognition methods are tightly related to geometrical concepts and numerous powerful tools have been developed during the last few decades, starting from the maximal likelihood method in the 1920's, to perceptrons in the 1960's, to kernel machines in the 1990's.

However, the geometric approach suffers from a major intrinsic limitation, which concerns the representational power of vectorial, feature-based descriptions. In fact, there are numerous application domains where either it is not possible to find satisfactory features or they are inefficient for learning purposes. This modeling difficulty typically occurs in cases when experts cannot define features in a straightforward way (e.g., protein descriptors vs. alignments), when data are high dimensional (e.g., images), when features consist of both numerical and categorical variables (e.g., person data, like weight, sex, eye color, etc.), and in the presence of missing or inhomogeneous data. But, probably, this situation arises most commonly when objects are described in terms of structural properties, such as parts and relations between parts, as is the case in shape recognition.

In the last few years, interest around purely similarity-based techniques has grown considerably. For example, within the supervised learning paradigm (where expert-labeled training data is assumed to be available) the now famous "kernel trick" shifts the focus from the choice of an appropriate set of features to the choice of a suitable kernel, which is related to object similarities. However, this shift of focus is only partial, as the classical interpretation of the notion of a kernel is that it provides an implicit transformation of the feature space rather than a purely similarity-based representation. Similarly, in the unsupervised domain, there has been an increasing interest around pairwise or even multiway algorithms, such as spectral and graph-theoretic clustering methods, which avoid the use of features altogether.

By departing from vector-space representations one is confronted with the challenging problem of dealing with (dis)similarities that do not necessarily possess the Euclidean behavior or not even obey the requirements of a metric. The lack of the Euclidean and/or metric properties undermines the very foundations of traditional pattern recognition theories and algorithms, and poses totally new theoretical/computational questions and challenges.

The aim of this workshop is to consolidate research efforts in this area, and to provide an informal discussion forum for researchers and practitioners interested in this important yet diverse subject. The discussion will revolve around two main themes, which basically correspond to the two fundamental questions that arise when abandoning the realm of vectorial, feature-based representations, namely:

  • How can one obtain suitable similarity information from data representations that are more powerful than, or simply different from, the vectorial?
  • How can similarity information be used in order to perform learning and classification tasks?

Accordingly, topics of interest include (but are not limited to):

  • Embedding and embeddability
  • Graph spectra and spectral geometry
  • Indefinite and structural kernels
  • Characterization of non-(geo)metric behaviour
  • Foundational issues
  • Measures of (geo)metric violations
  • Learning and combining similarities
  • Multiple-instance learning
  • Applications

Organizers

  • Joachim M. Buhmann, ETH Zurich, Switzerland
  • Robert P. W. Duin, Delft University of Technology, The Netherlands
  • Mario A. T. Figueiredo, Insituto Superior Tcnico, Lisbon, Portugal
  • Edwin R. Hancock, University of York, UK
  • Vittorio Murino, University of Verona, Italy
  • Marcello Pelillo, Ca' Foscari University, Venice, Italy (chair)

The primary goal of this workshop is to bring researchers working on various aspects of machine learning and games together. We want to provide a venue for discussing future directions for  machine learning in games, both for academia and the industry.

The intention is to keep the scope of the workshop rather broad and include topics such as:

  • Learning how to play games well, for games ranging from
    deterministic and discrete boardgames to non-deterministic,
    continuous, real time, action oriented games.
  • Player/opponent/team modeling, for goals such as improving
    artificial players in competitive games, mimicing human players, or game or learning curve adaptation.
  • Game analysis, for automatic skillranking, matchmaking, or player and team behavior analysis (fraud detection) in multiplayer games.
  • Automated content or story generation for games, possibly with attention to user specific constraints and preferences.
  • Game adaptivity, e.g. for raising or lowering difficulty levels
    dependent on the players proficiency, avoiding the emergence of player routines that are guaranteed to beat the game. This topic also includes concerns on game stability and performance guarantees for artificial opponents.
  • Novel learning scenarios arising from practical problems in games.
  • Machine learning perspectives in/from the games industry

Organizers

  • Christian Thurau, Fraunhofer IAIS and B-IT, University of Bonn
  • Kurt Driessens, Katholieke Universiteit Leuven
  • Olana Missura, Fraunhofer IAIS and University of Bonn

A central question in Computer Vision is how human prior knowledge can be integrated in order to build efficient and accurate learning systems. Computer Vision and Machine Learning Researchers have come up with many different models to incorporate their prior knowledge about the domain structure - such as probabilistic graphical models, conditional random fields, or structured support vector machines - which have been applied successfully to many different areas of application. The goal of this workshop is to bring together researchers from different directions of Computer Vision and Machine Learning, and to stimulate the discussion about the shared concepts, the recent progress and the remaining problems of learning and inference algorithms for structured representations in Computer Vision.

Program Committee

  • Yasemin Altun - MPI Tuebingen
  • Francis Bach - INRIA Paris
  • Matthew Blaschko - University of Oxford
  • Tiberio S. Caetano - NICTA
  • Daniel Cremers - TU Munich
  • Leo Grady - Siemens Research
  • Stefan Harmeling - MPI Tuebingen
  • Pawan Kumar - Stanford University
  • Pushmeet Kohli - Microsoft Research Cambridge
  • Bastian Leibe - RWTH Aachen
  • Simon Lucey - CMU
  • David McAllester - TTIC
  • Sebastian Nowozin - Microsoft Research Cambridge
  • Nikos Paragios - Ecole Centrale de Paris
  • Jan Peters - MPI Tuebingen
  • Deva Ramanan - UC Irvine
  • Bodo Rosenhahn - University of Hannover
  • Carsten Rother - Microsoft Research Cambridge
  • Bernhard Schoelkopf - MPI Tuebingen
  • Josef Sivic - INRIA Paris
  • Cristian Sminchisescu - University of Bonn
  • Ben Taskar - UPenn
  • Fernando de la Torre - CMU
  • Bill Triggs - INRS Grenoble
  • Jakob Verbeek - INRIA Grenoble
  • Louis Wehenkel - University of Liege
  • Richard Zemel - University of Toronto

Good decision-making is dependent on comprehensive, accurate knowledge. But the information relevant to many important decisions in areas such as business, government, medicine and scientific research is massive, and growing at an accelerating pace. Relevant raw data is widely available on the web and other data sources, but usually in order to be useful it must be gathered, extracted, organized, and normalized into a knowledge base.

Hand-built knowledge bases such as Wikipedia have made us all better decision-makers. However more than human editing will be necessary to create a wide variety of domain-specific, deeply comprehensive, more highly structured knowledge bases.

A variety of automated methods have begun to reach levels of accuracy and scalability that make them applicable to automatically constructing useful knowledge bases from text and other sources. These capabilities have been enabled by research in areas including natural language processing, information extraction, information integration, databases, search and machine learning. There are substantial scientific and engineering challenges in advancing and integrating such relevant methodologies.

This workshop gathered researchers in a variety of fields that contribute to the automated construction of knowledge bases.

There has recently been a tremendous amount of new work in this area, some of it in traditionally disconnected communities. In this workshop the organizers aim to bring these communities together.

Topics of interest include:

  • information extraction; open information extraction, named entity extraction; entity resolution, relation extraction.
  • information integration; schema alignment; ontology alignment; ontology constrution.
  • monolingual alignment, alignment between knowlege bases and text.
  • joint inference between text interpretation and knowledge base
  • pattern analysis, semantic analysis of natural language, reading the web, learning by reading.
  • databases; distributed information systems; probabilistic databases.
  • scalable computation; distributed computation.
  • information retrieval; search on mixtures of structured and unstructured data; querying under uncertainty.
  • machine learning; unsupervised, lightly-supervised and distantly-supervised learning; learning from naturally-available data.
    - human-computer collaboration in knowledge base construction; automated population of wikis.
    - dynamic data, online/on-the-fly adaptation of knowledge.
    - inference; scalable approximate inference.
    - languages, toolkits and systems for automated knowledge base construction.
    - demonstrations of existing automatically-built knowledge bases.

 

This special session aims to offer a meeting opportunity for academics and industry-related researchers, belonging to the various communities of Computational Intelligence, Machine Learning, Vision systems, Experimental Design and Data Mining to discuss new areas of active and autonomous learning, and to bridge the gap between data acquisition or experimentation and model building. How active sampling and data acquisition, can contribute towards the design and modelling of highly intelligent autonomous learning systems for the recognisition of complex (abnormal) behaviours is intended to be the catalyst and the aggregation stimulus for the overall event. Thus, there is a considerable interest in developing new methods and in extending and adapting existing traditional approaches. The results of the active learning competition will also be discussed in this special session.

Participation

Potential participants are invited to submit a paper to the special session on Active an Autonomous Learning. Please follow the regular submission guidelines of WCCI 2010 and submit your paper to the paper submission system. IMPORTANT: Select the correct special session: S111 Active and Autonomous Learning (AAL) at the bottom of the "Main Research Topic" menu AND notify the chairs of your submission by sending email to: aal@ clopinet . com.

Topics of interest to the workshop include (but are not limited to):

  • Experimental Design
  • Active Learning
  • Autonomous Learning
  • Incremental Learning
  • Autonomous intelligent systems
  • On-line learning
  • Machine Learning for Data Mining
  • Learning from unlabeled data
  • Artificial Vision
  • Agent and Multi-Agent Systems
  • Hybrid Systems
  • Unsupervised Learning
  • Classification Methods
  • Novelty Detection
  • Surveillance Systems and solutions (object tracking, multi-camera algorithms, behaviour analysis and learning, scene segmentation, system architecture aspects, operational procedures, usability, scalability)
  • Gesture and Posture Analysis and Recognition
  • Case Studies (shopping malls, railway stations, airport lounges, bank branches, etc)
  • Autonomous Robots
  • Industrial and Commercial Applications of Intelligent Methods
  • Biometric Identification and Recognition
  • Extraction of Biometric Features (fingerprint, iris, face, voice, palm, gait)
  • Email, Web and Networks Security

Participation in the active learning competition is not required to attend the workshop and vice versa. Competition participants can apply for travel support.

Session chairs

  • José García-Rodríguez (University of Alicante, Spain)
  • Isabelle Guyon (Clopinet Enterprises, USA)
  • Anastassia Angelopoulou (University of Westminster, UK)
  • Vincent Lemaire (Orange, France)

Organizing committee

  • Matthias Adankon (Ecole de technologie supérieure de Montréal, Canada)
  • Jorge Azorín (University of Alicante, Spain)
  • Alexis Bondu (EDF, France)
  • Marc Boullé (Orange, France)
  • Gavin Cawley (University of East Anglia, UK)
  • Olivier Chapelle (Yahhoo!, California)
  • Emilio Corchado (University of Burgos, Spain)
  • Juan Manuel Corchado (University of Salamanca, Spain)
  • Gideon Dror (Academic College of Tel-Aviv-Yaffo, Israel)
  • Emmanuel Faure (Institut des systèmes complexes, Paris, France)
  • Shaogang Gong (Queen Mary, University of London, UK)
  • Seiichi Ozawa (Kobe University, Japan)
  • Alexandra Psarrou (University of Westminster, UK)
  • Peter Roth (Graz University, Austria)
  • Asim Roy (Arizona State University, USA)
  • Amir Reza Saffari Azar (Graz University of Technology)
  • Fabrizio Smeraldi (Queen Mary, University of London, UK)
  • Alexander Statnikov (New York University, USA)

PAC-Bayes theory is a framework for deriving some of the tightest generalization bounds available. Many well established learning algorithms can be justified in the PAC-Bayes framework and even improved. PAC-Bayes bounds were originally applicable to classification, but over the last few years the theory has been extended to regression, density estimation, and problems with non iid data. The theory is well established within a small group of the statistical learning community, and has now matured to a level where it is relevant to a wider audience. The workshop will include tutorials on the foundations of the theory as well as recent findings through peer reviewed presentations.

Workshop topics

PAC Bayes theory or applications. In particular: application to:

  • regression
  • density estimation
  • hypothesis testing
  • structured density estimation
  • non-iid data
  • sequential data

Workshop Organisers

Hierarchies are becoming ever more popular for the organization of documents, particularly on the Web (Web directories are an example of such hierarchies). Along with their widespread use comes the need for automated classification of new documents to the categories in the hierarchy. As the size of the hierarchy grows and the number of documents to be classified increases, a number of interesting problems arise. In particular it is one of the rare situations where data sparsity remains an issue despite the vastness of available data. The reasons for this are the simultaneous increase in the number of classes and their hierarchical organization. The latter leads to a very high imbalance between the classes at different levels of the hierarchy. Additionally, the statistical dependence of the classes poses challenges and opportunities for the learning methods.

Research on large-scale classification so far has focused on situations involving a large number of documents and/or a large numbers of features, with a limited number of categories. However, this is not the case in hierarchical category systems, such as DMOZ, or the International Patent Classification, where in addition to the large number of documents and features, a large number of categories exist, in the order of tens or hundreds of thousands. Approaching this problem, either existing large-scale classifiers can be extended, or new methods need to be developed. The goal of this workshop is to discuss and assess some of these strategies, covering all or part of the issues mentioned above.

Organisers

  • Eric Gaussier, LIG, Grenoble, France
  • George Paliouras, NCSR "Demokritos", Athens, Greece
  • Aris Kosmopoulos, NCSR "Demokritos" and AUEB, Athens, Greece
    Sujeevan Aseervatham, LIG, Grenoble, France

The workshop shall serve two purposes. On the one hand, we will have invited speakers who present talks on multiple testing methodology and practice in a tutorial style for an audience that is familiar with basic concepts of statistical testing, but has no particular expertise in MCPs. Directly after such a theoretical talk, we will present how the things which have just been presented can be realized in our new software (this practical presentation will be done by one of the coding team members).   In a separate time slot, contributed posters will be presented.

 

This symposium addresses a topic that has spurred vigorous scientific debate of late in the fields of neuroscience and machine learning: causality in time-series data. In neuroscience, causal inference in brain signal activity (EEG, MEG, fMRI, etc.) is challenged by relatively rough prior knowledge of brain connectivity and by sensor limitations (mixing of sources). On the machine learning side, as the Causality workshop last year’s NIPS conference has evidenced for static (non-time series) data, there are issues of whether or not graphical models (directed acyclic graphs) pioneered by Judea Pearl, Peter Spirtes, and others can reliably provide a cornerstone of causal inference, whereas in neuroscience there are issues of whether Granger type causality inference is appropriate given the source mixing problem, traditionally addressed by ICA methods. Further topics, yet to be fully explored, are non-linearity, non-Gaussianity and full causal graph inference in high-dimensional time series data. Many ideas in causality research have been developed by and are of direct interest and relevance to researchers from fields beyond ML and neuroscience: economics (i.e. the Nobel Prize winning work of the late Clive Granger, which we will pay tribute to), process and controls engineering, sociology, etc. Despite the long-standing challenges of time-series causality, both theoretical and computational, the recent emergence of cornerstone developments and efficient computational learning methods all point to the likely growth of activity in this seminal topic.

Along with the stimulating discussion of recent research on time-series causality, we will present and highlight time-series datasets added to the Causality Workbench, which have grown out of last year’s Causality challenge and NIPS workshop, some of which are neuroscience related.

Programme Committee

  • Luiz Baccala (Escola Politecnica da Universidade de Sao Paulo, Brazil)
  • Katarina Blinowska (University of Warsaw, Poland)
  • Alessio Moneta (Max Planck Institute of Economics, Germany)
  • Mischa Rosenblum (Potsdam University, Germany)
  • Bjoern Schelter (Freiburg Center for Data Analysis and Modeling, Germany)
  • Pedro Valdes-Sosa (Neurosciences Center of Cuba)

Nowadays, there are massive amounts of heterogeneous electronic information available on the Web. People with disabilities, among other groups potentially in uenced by the digital gap, face great barriers when trying to access information. Sometimes their disability makes their interaction the ICT environment (eg., computers, mobile phones, multimedia players and other hardware devices) more dicult. Furthermore, the contents are delivered in such formats that cannot be accessed by people with disability and the elderly. The challenge for their complete integration in information society has to be analyzed from di erent technology approaches.

Recent developments in Machine Learning are improving the way people with disabilities access to digital information resources. From the hardware perspective, Machine Learning can be a core part for the correct design of accessible interaction systems of such users with computers (such as BCI). From the contents perspective, Machine Learning can provide tools to adapt contents (for instance changing the modality in which it is accessed) to users with special needs. From the users' perspective, Machine Learning can help constructing a good user modeling, as well as the particular context in which the information is accessed.

Logo

Follow us:

© Knowledge 4 All Foundation Ltd.