Zulu Challenge

http://labh-curien.univ-st-etienne.fr/zulu/
Supported by Pascal 2 Network of Excellence

Keywords: Active learning, grammatical inference, DFA and grammars

Abstract:

Zulu is an active learning competition. Participants are to build
algorithms that can learn deterministic finite automata (DFA) by making
the smallest number of membership queries to the server/oracle.

Motivations:

When learning language models, techniques usually make use of huge corpora
that are unavailable in many less resourced languages (such as the Zulu
language). One possible way around this problem is to interrogate an
expert with a number of chosen queries, in an interactive mode, until a
satisfying language model is reached. In this case, an important indicator
of success is the amount of energy the expert has spent in order for
learning to be successful. A nice learning paradigm covering this
situation is that of Query Learning, introduced by Dana Angluin.

In the field of Grammatical Inference, Query Learning was thoroughly
investigated to learn deterministic finite automata (DFA). As negative
results, it was proved that DFA could not be learned from just a
polynomial number of membership queries nor from just a polynomial number
of strong equivalence queries. On the other hand, algorithm L* designed by
Angluin, was proved to learn DFA from a polynomial number of both
membership and equivalence queries. These results yield several
successfull applications in Robotics, Games and Agents Technologies,
Information Retrieval, Hardware and Software Verification.

However, what has not been hardly studied is how to optimise the learning
task by trying to minimize the number of queries while making queries for
which the Oracle’s work and answers are simple. These are strong
motivations for stemming research in the direction of developing new
interactive learning strategies and algorithms, that is the aim of this
competition.

The competition:

Zulu (http://labh-curien.univ-st-etienne.fr/zulu/) is both a web based
platform simulating an Oracle in a DFA learning task and a competition.

As a web platform, Zulu allows users to generate tasks, to interact with
the Oracle in learning sessions and to record the results of the users. It
provides the users with a baseline algorithm written in JAVA, or the
elements allowing to build from scratch a new learning algorithm capable
of interacting with the server.

The server can be accessed by any user/learner who has opened an account.
The server acts as an Oracle for membership queries. A player can log in
and ask for a target DFA. The server then computes how many queries it
needs to learn a reasonable machine (reasonable means less than 30%
classification errors), and invites the player to interact in a learning
session in which he can ask up to that number of queries.

At the end of the learning process the server gives the learner a set of
unlabelled strings (a test set). The labels the learner submits are used
to compute his score.

As a starting point the baseline algorithm, which is a simple variation of
L*, with some sampling done to simulate equivalence queries, is given to
the user, who can therefore play with some simple JAVA code for a start.

The competition itself will be held in the spring of 2010 and the results
will be presented during a special session at the International Colloquium
on Grammatical Inference in Valencia, Spain, September 13-16, 2010
(http://users.dsic.upv.es/workshops/icgi2010/)

Schedule:

* from now to March 1st, 2010: Zulu platform is open, anyone may
register and have fun
* March 1st: Official beginning of the competition
* May 15th: Deadline for scoring, submissions closed
* June 1st: Notifications of the results
* June 20th: Deadline for submission of abstracts explaining
participants strategies
* September 13-16th: workshop at ICGI

Prizes and publications:
The winner of the Zulu competition will receive a prize, to be announced
on the Zulu webpage. Participants are encouraged to present their
innovations either as full papers to the ICGI 2010 conference, or as
extended abstracts to the Zulu workshop that will be organised during
ICGI. A journal special issue will also be considered.

Scientific committee:
* Dana Angluin, Yale University, USA
* Leo Becerra Bonache, Universidad de Tarragona, Spain
* François Coste, IRISA, Rennes, France
* Alex Clark, Royal Holloway University of London, UK
* Ricard Gavalda, Universidad Politecnica de Barcelona, Spain
* Colin de la Higuera, University of Nantes, France
* Jean-Christophe Janodet, University of Lyon, France
* Aurelien Lemay, University of Lille, France
* Laurent Miclet, ENSAT Lannion and IRISA, France
* Tim Oates, University of Maryland, USA
* Anssi Yli Jyra, University of Helsinki, Finland
* Menno van Zaanen, Tilburg University, The Netherlands

Two positions at the Australian National University

The Australian National University has advertised two tenure-track style
positions in Artificial Intelligence (including Machine Learning).

The positions are research and teaching, but with a reduced teaching load for an
initial period. Closing date 14 February 2010. Details here:
http://jobs.anu.edu.au/PositionDetail.aspx?p=1043

COLT 2010 CFP

COLT 2010 – Call for Papers

The 23rd Annual Conference on Learning Theory (COLT 2010) will take
place in Haifa, Israel, on June 27-29, 2010 and will be co-located
with ICML 2010. We invite submissions of papers addressing theoretical
aspects of machine learning and empirical inference. We strongly
support a broad definition of learning theory, including:

* Analysis of learning algorithms and their generalization ability
* Computational complexity of learning
* Bayesian analysis
* Statistical mechanics of learning systems
* Optimization procedures for learning
* Kernel methods
* Inductive inference
* Boolean function learning
* Unsupervised and semi-supervised learning and clustering
* On-line learning and relative loss bounds
* Learning in planning and control, including reinforcement learning
* Learning in games, multi-agent learning
* Mathematical analysis of learning in related fields, e.g., game
theory, natural language processing, neuroscience, bioinformatics,
privacy and security, machine vision, data mining, information retrieval

We are also interested in papers that include viewpoints that are new
to the COLT community. We welcome experimental and algorithmic papers
provided they are relevant to the focus of the conference by
elucidating theoretical results in learning. Also, while the primary
focus of the conference is theoretical, papers can be strengthened by
the inclusion of relevant experimental results.

Papers that have previously appeared in journals or at other
conferences, or that are being submitted to other conferences, are not
appropriate for COLT. Papers that include work that has already been
submitted for journal publication may be submitted to COLT, as long as
the papers have not been accepted for publication by the COLT
submission deadline (conditionally or otherwise) and that the paper is
not expected to be published before the COLT conference (June 2010).

Feedback on Review Quality
There will be no rebuttal phase this year. However, authors will be
given the opportunity to assess the quality of reviews and provide
feedback to the reviewers, after the decisions have been made. These
assessments will be used in particular to determine the Best Reviewer
award (see below).

Paper and Reviewer Awards
This year, COLT will award both best paper and best student paper
awards. Best student papers must be authored or coauthored by a
student. Authors must indicate at submission time if they wish their
paper to be eligible for a student award. This does not preclude the
paper to be eligible for the best paper award.

To further emphasize the importance of the reviewing quality, this
year, COLT will also award a best reviewer award to the reviewer who
has provided the most insightful and useful comments.

Open Problems Session
We also invite submission of open problems (see separate call). These
should be constrained to two pages. There is a shorter reviewing
period for the open problems. Accepted contributions will be allocated
short presentation slots in a special open problems session and will
be allowed two pages each in the proceedings.

Paper Format and Electronic Submission Instructions
Formatting and submission instructions will be available in early
December at the conference website. Submissions should include the
title, authors’ names, and a 200-word summary of the paper suitable
for the conference program. Papers should not exceed 13 pages
(including bibliography) and should be formatted according to the
following style file and sample LaTeX source (colt10e.sty, colt10-
sample.tar.gz). Authors not using latex should ensure that their
document complies with similar formatting (similar margins, 11pt font,
single column). Shorter papers are strongly encouraged. Additional
material beyond the 13 page limit can be placed in the appendix and
might be read, at the discretion of the program committee.

Important Dates
Preliminary call for papers issued October 15,2009
Electronic submission of papers (due by 5:59pm PST) February 19, 2010
Electronic submission of open problems March 13, 2010
Notice of acceptance or rejection May 07, 2010
Submission of final version May 21, 2010
Feedback on reviews due May 28, 2010
Joint ICML/COLT workshop day June 25, 2010
2010 COLT conference June 27-29, 2010

OrganizationProgram Co-chairs:
* Adam Tauman Kalai (Microsoft Research)
* Mehryar Mohri (Courant Institute of Mathematical Sciences and Google Research)

Program Committee:
Shivani Agarwal Mikhail Belkin
Shai Ben-David Nicolò Cesa-Bianchi
Ofer Dekel Steve Hanneke
Jeff Jackson Sham Kakade
Vladimir Koltchinskii Katrina Ligett
Phil Long Gabor Lugosi
Ulrike von Luxburg Yishay Mansour
Ryan O’Donnell Massimiliano Pontil
Robert Schapire Rocco Servedio
Shai Shalev-Shwartz John Shawe-Taylor
Gilles Stoltz Ambuj Tewari
Jenn Wortman Vaughan Santosh Vempala
Manfred Warmuth Robert Williamson
Thomas Zeugmann Tong Zhang

Publicity Chair:
* Sandra Zilles (University of Regina)

Local Arrangements Chair:
* Shai Fine (IBM Research Haifa)

2-year postdoc in Machine Learning and Reverse-Modeling of biological networks (Evry, near Paris, France)

A 2-YEAR POSTDOC POSITION AVAILABLE in MACHINE LEARNING and REVERSE-MODELING OF SIGNALING and GENE REGULATORY NETWORKS at IBISC (University of Evry – GENOPOLE, France)

The Machine Learning, Modeling and Data Integration Group group at laboratory IBISC at University of Evry – Genopole (30km Paris, France) is looking for postdoctoral candidates for a two-year position with a strong background in statistical learning (graphical models and kernel methods) and an interest for interdisciplinary research aimed towards reverse-modeling of biological networks.

The successful applicant will participate to an ANR project called ODESSA that began on December 1rst, 2009, in collaboration with the l’Institut de Génomique Fonctionnelle de Lyon (IGFL, Ecole Normale Supérieure de Lyon, France). He/she will investigate new algorithms for structure and module learning in nonlinear state-space models. A sound knowledge in graphical models (dynamical models, mixture models, latent variable models) and kernel-based approaches is asked. The targeted application is the identification of signaling and gene regulatory networks involved in the cellular response to retinoic acid from large scale data (next generation sequencers). Experience in large scale data-mining of post-genomic data will be appreciated but is not required.

The successful candidate will have a Ph.D. in Computer Science, Mathematics or Bioinformatics with a strong publication record.

Applications for this role should include a statement of research experience and interests, a CV, two reference letter(s) from appropriate academic sources and the two best publications.

Please send your applications in electronic form to Prof. Florence d’Alché-Buc : florence.dalche (at) ibisc.univ-evry.fr

The position can start in early 2010 and not later than June 2010.

Deadline for the first round of applications is Jan, 31 2010.

More details available at http://amisbio.ibisc.fr/opencms/opencms/amis/en/jobs/

PhD student for computational genetics

Please see below link for details of a vacancy for a PhD student who will further develop Bayesian methods for use in the analysis of genome-wide association studies:

http://www.ru.nl/vacaturedetails?recid=497857

If somebody has a good candidate, we would like to ask him/her to forward this advert to this person.

2 inderdiscplinary postdoc positions in London

Research Associate – Capitalizing on Gravitational Shear

A brief summary and pointer to the formal ad can be viewed on
cosmocoffee http://cosmocoffee.info/viewtopic.php?p=4369#4369

Workshop ECIR 2010 on Large-Scale Hierarchical Classification

Large-Scale Hierarchical Classification Workshop European Conference on
Information Retrieval 2010

March 28, 2010
The Open University in Milton Keynes, UK

http://lshtc.iit.demokritos.gr/workshop

*********************************************************
THE CHALLENGE

The workshop is associated to the PASCAL 2 Large-Scale Hierarchical Text
Classification Challenge. The challenge was completed recently with the very
active participation of several teams from around the world. The results are
publicly available at: http://lshtc.iit.demokritos.gr/

*********************************************************
SCOPE

Hierarchies are becoming ever more popular for the organization of
documents, particularly on the Web (e.g. Web directories). Along with their
widespread use comes the need for automated classification of new documents
to the categories in the hierarchy. Research on large-scale classification
so far has focused on large numbers of documents and/or large numbers of
features, with a limited number of categories. However, this is not the case
in hierarchical category systems, such as DMOZ. 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. In particular some of the issues that we expect to cover
in the workshop are:
* Learning to classify against many categories.
* Data sparseness in the presence of large datasets.
* Use of the statistical dependence of hierarchically organized classes.
* The role of shrinkage methods in large hierarchies.
* Ensemble methods for hierarchical classification.
* Extending existing large-scale classifiers to hierarchies.
* Challenging hierarchical classification tasks and datasets.

*********************************************************
IMPORTANT DATES

* Paper submission – January 18
* Acceptance notification – February 15
* Final paper – March 1

*********************************************************
SUBMISSIONS

We expect submissions by the various teams who have participated to the
LSHTC challenge. At the same time, we strongly encourage submissions from
researchers not participating to the challenge. The Easychair electronic
submission system will be used for the papers. Please, refer to the workshop
page for details about the submission format and process.

*********************************************************
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

—

Research Readership in Computer Science

Oxford Brookes University – School of Technology
Starting salary: £46,509 rising annually to £52,346

negligible teaching activities!
see

http://www.jobs.ac.uk/job/AAF163/reader-in-computer-science

Email philiptorr(at)brookes.ac.uk if any questions

Engineer and Post-doc positions – 3D avatar for Image Registration

Engineer and Post-doc positions – 3D avatar for Image Registration

The Inselspital, University Hospital Bern (www.insel.ch) is starting a KTI funded project with the University of Geneva (www.miralab.ch) and Nhumi Technologies GmbH (www.nhumi.com) as business partner to design a new system for navigating healthcare data using a 3D model of the human anatomy. The purpose is to enable patients and clinicians to see on a 3D avatar where images (e.g. X-rays, MRI’s) have been taken and therefore to request a relevant image in a single click.

For this project, the Inselspital is looking for an experienced engineer or a post-doc with strong software development skills who would have the responsibility to design, implement and test innovative techniques to align or register a set of real medical images with artificially generated images from a standard model of the human anatomy. The candidate will be integrated in the team of the Inselspital and will also work closely with the development lab of Nhumi Technologies and should be at ease with both academic and industrial environment.

The candidate will be offered a competitive salary for two years and will have the opportunity to put his ideas at work in a challenging environment. As a member of a dynamic team, she/he will be asked to work independently and to communicate regularly about her/his activities.

Required skills:

* English speaking (German a plus)
* Literate in Java and strong programming skills. Knowledge of web services and javascript.
* Mathematical background or experience in Computational Geometry algorithms or 3D Graphics programming.
* Master of Science or PhD degree in Computer Science, Mathematics, Computer Graphics or related fields.
* Experience in image registration and medical informatics a plus.

Job location:

* Inselspital, Bern
* Nhumi Technologies, Zurich

Job duration:
2 years minimum

For more information, please do not hesitate to contact

Dr. med Michael A. Patak, Oberarzt
Institut für Diagnostische, Interventionelle und Pädiatrische Radiologie
Inselspital, Universitätsspital Bern
Freiburgstrasse
3010 Bern

michael.patak(at)insel.ch

Postdoc positions in SMT and/or CLIR, Xerox Research Centre Europe

The Cross-Language Technologies (CLT) area of the Xerox Research Centre Europe (XRCE) conducts research in the following areas: Statistical Machine Translation; Multilingual Terminology Extraction; Cross-Language Information Retrieval (CLIR), Categorization and Clustering. The team is part of the PASCAL 2 European Network of Excellence, ensuring a strong network of academic collaborations.

We are opening 2 Postdoc positions for researchers interested in the areas of Statistical Machine Translation and/or CLIR. For the first position, we are looking for a candidate with a strong Machine Learning background, for the second position for a candidate with a strong SMT, CLIR, or NLP background.

Required experience and qualifications:

For the first position:

* PhD in computer science, statistics or mathematics with focus on Machine Learning.

* Acquaintance with SMT or NLP is a plus.

For the second position:

* PhD in computer science or computational linguistics with focus on SMT, CLIR, or statistical NLP.

* Acquaintance with Machine Learning is a plus.

For both positions:

* Good programming skills in C, C++, and/or Python.

* A good command of English is required, as well as open-mindedness and the will to collaborate within a team.

Starting date: Beginning of 2010
Duration: 18 months

Application instructions: please email your CV and covering letter, with message subject containing the expression “[CLT Postdocs 2010]” to xrce-candidates and to Nicola.Cancedda at xrce.xerox.com.

More details at:

http://www.xrce.xerox.com/About-XRCE/Jobs/Two-Postdoc-Positions-Cross-Language-Technologies .

Inquiries can be sent to Nicola.Cancedda at xrce.xerox.com.

XRCE is a highly innovative place and we strongly encourage publication and interaction with the scientific community.