Last call for Challenge Proposals! Deadline 3rd February 2012

Next Call for Challenge Proposals – Winter 2012
Deadline, Feb. 3rd, 2012

The goal of Pascal-II Challenges is to advance the state of the art about cognitive systems. The cognitive banner is large: welcome are proposals related to

* communication (language, voice, vision, haptics, brain signals, representation, feature construction, – non exhaustive list)
* interaction (games, robotics, social networks, preferences, validation, experiment design, distributed/decentralized decision – n.e.l.)
* applicative settings (multi-task and transfert, drift, privacy and anonymization – n.e.l.)
* theoretical and/or algorithmic aspects of communication, interaction, computational thinking, and cognition at large.

Challenges come in three possible flavors:

* Applicative challenges aim at advancing the state of the art in some real-world domain; they should come with real-world, comprehensive datasets, with different levels of difficulty if possible.
* Exploratory challenges aim at investigating a new field/mode of ML; they should come with highly flexible settings, in order to understand where the difficulty is and to perform lesion studies.
* Theoretical challenges aim at long term research; they should provide (experimental or theoretical) milestones for the road.

NB: A legacy of Pascal-I is a perennial repository, storing all previous challenge datasets; Pascal-II challenges will participate to this repository, contributing to a principled, well-recognized and up-to-date benchmarking methodology.

PostDoc in Dimensionality Reduction and Information Visualization at UCLouvain

The Université catholique de Louvain invites applications for a 2-year postdoctoral position in Machine Learning/Information Visualisation, beginning July 1, 2013.

The visual interpretation of data is an essential step to guide any further processing or decision making. Data visualization is tackled independently from two different angles in the scientific community. The domain of machine learning addresses mainly statistical, mathematical, and algorithmic aspects with dimensionality reduction (DR) techniques, whereas the field of information visualization focuses on the interaction with the user (man-computer interface, visual efficacy, user-friendliness). In this context, the project aims at bridging the two approaches. More specifically, the project intends to import the concepts of interactivity and controllability from the field of information visualization and to integrate them in advanced DR techniques in order to improve their acceptance by users and broaden their range of application.

The successful applicant will hold a Ph.D. degree delivered not earlier than July 1, 2007, and must not have the Belgian nationality. In addition, the candidate must not have lived in Belgium for more than 2 years since July 1, 2010. The net salary after deduction of taxes and social security is about 26000 euros per year, or more, depending on seniority.

Applicants should be knowledgeable in machine learning, data mining, information visualisation, with a particular interest in manifold learning, dimensionality reduction and information retrieval. Applicants should have a strong background in computer science and applied mathematics.

A working knowledge of English language is mandatory. French is an optional asset.

Working location will be Louvain-la-Neuve, a lively pedestrian town in the suburbs of Brussels, where most of the Université catholique de Louvain is located. The work will be carried out in collaboration with Profs. John A. Lee
( and Michel Verleysen (, in the ICTEAM institute (

Application procedure: Interested individuals should send a CV, a brief statement of research and development interests (max. 1 page), and the names and contact details of two references by e-mail to John Lee
( with subject “Postdoc DRedVis”.

Candidates interested should send their application before March 15th, 2013; we reserve the right to accept late applications. The position will be available for an initial period of 1 year, with possible one-year extension after mid-term evaluation.

Postdoctoral Position in Machine Learning at INRIA Lille – Team SequeL

Applications are invited for a Postdoctoral position on the general area of “Sequential Decision-making in Online Marketing” at INRIA Lille – Team SequeL. Below is the detail of this call.

Title: Sequential Decision-making in Online Marketing: Optimizing the Lifetime Value of Customers

Keywords: sequential decision-making, reinforcement learning, online marketing and advertising, exploration/exploitation dilemma, bandit algorithms, adaptive resource allocation, regret minimization, optimization

Research Program:

The candidate is expected to conduct research on both theoretical and applied aspects of the problem of “Sequential Decision-making in Online Marketing” and related problems (see the description below), collaborate with researchers and Ph.D. students at INRIA and outside, and publish the results of her/his research in conferences and journals. The candidate will work with Mohammad Ghavamzadeh ( and other researchers at Team SequeL (

With the growth of online marketing, customers visit websites on a regular basis (sometimes daily in the case of banking, e-commerce, and media websites), and at each visit a stream of interactions occurs between the company (promotion, advertisement, email) and a customer (purchase, click on an ad, signing up for a newsletter). This creates many opportunities for a company to reach a customer and make decisions to optimize its objective function (revenue, customer satisfaction, etc.). Today, these marketing decisions are mainly made in a myopic way (mainly using contextual bandit algorithms) without taking into account the lifetime value of the customer. This myopic approach assumes that a decision made by the company does not affect the customer’s future interactions with the company. However, in many applications the sequential nature of the problem is significant (having a long-term relation with customers is important for the company), and thus, myopic decisions have po!
or performance in these problems. This creates an opportunity for reinforcement learning (RL) techniques to play a significant role in this emerging field.

The objective of this research program is to answer fundamental questions related to the use of myopic (contextual bandit algorithms) and non-myopic (sequential decision-making and RL algorithms) decision-making methods in the growing field of online marketing. Questions such as

– Feature selection and dealing with high dimensional data: discovering the right representation for the problem at hand and dealing with the size and dimensionality of the data are among the most important questions in these applications. The size and dimensionality of the data create difficulties for the standard sequential decision-making algorithms. This is closely related to another growing research direction: sequential decision-making with big data.

– Off-policy evaluation: how to evaluate a policy learned from a batch of historical data generated with a different policy (usually the company’s policy) with minimum interaction with the real-world environment. Running a strategy on the real system can be costly: it usually takes a long time to have a reasonable evaluation of its quality and more importantly is the risk of a big loss in case the strategy is not good.

– Discovering patterns in the sales funnel in order to find strategies to direct more customers through the funnel to the final sale.

– Dealing with the non-stationarity, mainly caused by change in the preferences of the customer, arrival and departure of customers, evolution of webpage contents, etc., and delayed feedback (significant delay between an action taken by the marketer and its effects on the customer) in the online marketing applications.


The applicant will have a Ph.D. degree (by the starting date of the postdoctoral position) in Computer Science, Statistics, or related fields, with background in reinforcement learning, bandit algorithms, statistics, and optimization. Programming skills will be considered as a plus. The working language of the group is English, so the candidate is expected to have good communication skills in English.

About INRIA and Team SequeL:

SequeL ( is one of the most dynamic teams at INRIA (, with over 25 researchers and Ph.D. students working on several aspects of machine learning from theory to application, including statistical learning, reinforcement learning, and sequential decision-making. The SequeL team is involved in national and European research projects and has collaboration with international research groups. This allows the postdoctoral candidate to collaborate with leading researchers in the field at top universities in Europe and North America such as University College of London (UCL), University of Alberta, and McGill University. Moreover, in this project there is the possibility of close collaboration with an online marketing company in the US. Lille is the capital of the north of France, a metropolis with over one million inhabitants, and with excellent train connection to Brussels (30min), Paris (1h) and London (1h30).


– Duration: 16 months – starting date of the contract : November 1, 2013
– Salary: 2620.84 Euros gross/month monthly salary
– Monthly salary after taxes: around 2138 Euros (medical insurance included)
– Possibility of French courses
– Help for housing
– Participation for transportation
– Scientific Resident card and help for husband/wife visa

Application Submission:

The application should include a brief description of the applicant’s research interests and past experience, plus a CV that contains her/his degrees, GPAs, relevant publications, name and contact information of up to three references, and other relevant documents. Please send your application to The deadline for the application is April 15 but the applicants are encouraged to submit their application as soon as possible.

This call has also been posted on

1) my webpage at

2) the INRIA website at:

PhD Studentship in statistical machine translation – Xerox & University of Grenoble

PhD student, Statistical Machine Translation in Grenoble (France)

The Machine Learning for Document Access and Translation group of the Xerox Research Centre Europe (XRCE) conducts research in Statistical Machine Translation and Information Retrieval, Categorization and Clustering using advanced machine learning methods.

We are opening a position for a PhD studentship in the field of Statistical Machine Translation. The ideal candidate has a strong undergraduate curriculum in SMT, or otherwise in statistical NLP, and is passionate about devising, implementing and evaluating effective solutions to challenging language problems.

This PhD will be co-supervised with the Laboratory of Informatics of Grenoble (LIG) which conducts research in Machine Translation and Natural Language Processing. The candidate will register as a graduate PhD student at University of Grenoble.

Required experience and qualifications:

– Master-level degree in computer science, mathematics, statistics, or computational linguistics with focus on Machine Translation, statistical NLP, or machine Learning.
– A good command of English, as well as open-mindedness and the will to collaborate within a team.

Additional desirable features:

– A publication record
– Evidence of implementing systems.

Preferred starting date: ASAP

Application instructions

Inquiries can be sent to
Nicola.Cancedda at
Laurent.Besacier at

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

Postdoctoral Research Fellow on “Low-complexity source separation algorithms”, CVSSP, University of Surrey, U.K. (Closing on March 17th, 2013)

Research Fellow

Low-complexity source separation algorithms

Centre for Vision Speech and Signal Processing (CVSSP)
University of Surrey, United Kingdom
Salary: £29,541-£30,424 per annum
(Subject to qualifications and experience)
Applications are invited for a three-year postdoctoral research fellow position available at CVSSP, starting on Monday, April 1, 2013, to work on a project entitled “Signal Processing Solutions for a Networked Battlespace”, funded by the Engineering and Physical Sciences Research Council (EPSRC) and Defence Science and Technology Laboratory (Dstl), as part of the Ministry of Defence (MoD) University Defence Research Centre (UDRC) Scheme in signal processing. This project will be undertaken by a unique consortium of academic experts from Loughborough, Surrey, Strathclyde and Cardiff (LSSC) Universities together with six industrial project partners QinetiQ, Selex-Galileo, Thales, Texas Instruments, PrismTech and Steepest Ascent. The overall aim of the project is to provide fundamental signal processing solutions to enable intelligent and robust processing of the very large amount of multi-sensor data acquired from various networked communications and weapons platforms, in order to retain military advantage and mitigate smart adversaries who present multiple threats within an anarchic and extended operating area (battlespace). The research fellow will be expected to work in close collaboration with our academic and industrial partners together with members of the lead consortium based at Edinburgh and Heriott Watt Universities.
The prospective research fellow will be expected to develop low-complexity robust algorithms for underdetermined, convolutive signal separation, broadband distributed beamforming. The work will be facilitated by low-rank and sparse representations, and directed toward fast implementations. He/she will develop robust source separation algorithms in highly dense signal environments, with the presence of uncertainties, such as weak signals and the unknown number of targets.
Successful applicants will join the CVSSP, a leading research group in sensory (visual and auditory) data analysis and interpretation, and will work closely with Dr Wenwu Wang, Prof Josef Kittler and Dr Philip Jackson. CVSSP is one of the largest UK research groups in machine vision and audition with more than 120 researchers, with core expertise in Signal Processing, Image and Video Processing, Pattern Recognition, Computer Vision, Machine Learning and Artificial Intelligence, Computer Graphics and Human Computer Interaction. CVSSP forms part of the Department of Electronic Engineering, which received one of the highest ratings (joint second position across the UK) in the last research quality assessment, i.e. 2008 RAE, with 70% of its research classified as either 4* (“world-leading”) or 3* (“internationally excellent”).

Applicants should have a PhD degree or equivalent in electrical and electronic engineering, computer science, mathematical science, statistics, physics, or related disciplines. Applicants should be able to demonstrate excellent mathematical, analytical and computer programming skills. Advantages will be given to the applicants who have experience in sparse representations, blind source separation, low-rank linear algebra, and/or machine learning.
For informal inquiries about the position, please contact Dr Wenwu Wang (
For an application pack and to apply on-line please go to our website: If you are unable to apply on-line please contact Mr Peter Li, HR Assistant on Tel: +44 (0) 1483 683419 or email:
The closing date for applications is Sunday March 17th, 2013.

For further information about the University of Surrey, please visit

We acknowledge, understand and embrace cultural diversity.

ICML ’13 Workshop – Call for papers

ICML ’13 Workshop – Call for papers:
Prediction with Sequential Models
June 21st or 22nd, Atlanta, GA, USA
*** Website: ***


Supervised and unsupervised function learning is a vast domain with a plethora of standard algorithmic solutions. Most of these methods learn a monolithic predictor function in the sense that each test instance is processed in a single-step, atomic process. In contrast, some recent studies have proposed a different paradigm in which *prediction is reformulated as a sequential decision process* and *learning the predictor function corresponds to solving a dynamic control problem*. These new approaches bridge “classical” supervised and unsupervised learning problems with the fields of control theory and reinforcement learning (RL), and raise interesting questions on different domains ranging from reinforcement learning to representation learning.

This workshop aims at gathering the various machine learning sub-communities that have worked around the subject and discuss the aforementioned issues. The topics of interest include, but are not limited to:
• Generic topics:
– Classification, ranking,
– Budgeted and/or cost-sensitive classification
– Structured prediction
– Sparse coding with sequential models
– Feature selection.
• Reinforcement learning applied to learning sequential functions:
– RL with many discrete actions
– RL in high-dimensional spaces
– Inverse RL
• Applications:
– Real-time detection and classification
– Text/image classification and information extraction
– Trigger design in high-energy particle physics
– Web-page ranking
– Medical diagnosis
Program Committee
• Francis Bach – Laboratoire d’Informatique de l’Ecole Normale Superieure – INRIA-Sierra – France
• Aaron Courville – University of Montreal – Canada
• Jason Eisner – Johns Hopkins University – USA
• Damien Ernst – University of Liege – Belgium
• Hugo Larochelle – University of Sherbrooke – Canada
• Francis Maes – Catholic University of Leuven – Belgium
• Rémi Munos – INRIA Sequel – France
• Philippe Preux – University Lille 3 – INRIA Sequel – France
• Thomas Rückstieß – Technische Universitat Munchen – Germany
• Csaba Szepesvári – University of Alberta – Canada
• Kilian Weinberger – Washington University – USA

• Djalel Benbouzid (Paris Sud – CNRS)
• Ludovic Denoyer (UPMC – LIP6)
• Gabriel Dulac-Arnold (UPMC – LIP6)
• Patrick Gallinari (UPMC – LIP6)
• Balàzs Kégl (Paris Sud – CNRS)
• Michèle Sébag (Paris Sud – CNRS)

PhD studentship in evolutionary theory at Royal Holloway, U. of London

Intra-genomic Conflict and Medical Disorders

Dr Francisco Ubeda and Prof. Vincent A.A. Jansen

Research Outline
Intra-genomic conflict defies the logic of natural selection: why would natural selection favor any gene whose expression reduces the fitness of its host? However intra-genomic conflict has left its signature in many molecular mechanisms. A paradigmatic example of evolution driven by intra-genomic conflict is the case of genomic imprinting where conflict between paternally inherited and maternally inherited genes in the same individual results in silencing of one gene but not the other (1).

Recently, genomic imprinting (and intra-genomic conflict in general) has been linked to several diseases (2). For example, deletion of the PWS/AS cluster of imprinted genes causes Prader-Willi syndrome (PWS) when the deletion is paternally inherited but Angelman syndrome (AS) when it is maternally inherited (3). The clinical phenotype, regarding appetite and activity levels, of children suffering from these syndromes is the reverse: poor sucking and low weight in children with PWS but insatiable appetite and obesity in children with AS (3).

This intriguing reversal of the clinical phenotype of a deletion is best explained in the light of conflict between genes with different parental origin. In particular, it can be explained when paternally inherited copies favor a greater allocation of maternal resources to offspring than the maternally inherited copy does (4). We are interested in further exploring the role of intra-genomic conflict in disease. Can we predict the risk of developing diseases caused by genes in conflict? Can we suggest epigenetic modifications that may palliate some symptoms?

In this project we will formulate mathematical models for the evolution of intra-genomic conflict and make specific predictions about the outcomes. We will test the predictions of our models against the medical literature. This research will require a trans-disciplinary approach that uses mathematical and computational models to synthesize the fields of molecular biology, genetics, medicine, evolutionary biology, and behavioral ecology. We hope to apply this approach to understand the evolution of genomic imprinting, sex-determination, and disease virulence among others.

This project is suitable for candidates with some background or experience in mathematical modeling or simulation at undergraduate level. We are looking for candidates, either with a background in the life sciences, and experience in mathematical or simulation modeling, or for candidates with a background in a quantitative subject (e.g. mathematics, computer science, physics) and an affinity for research in ecology and evolution.

The studentship will be held in the School of Biological Sciences of Royal Holloway, University of London. The research in the School covers the breadth of biology and hosts a number of theoretical researchers. The School was ranked among the best UK Bioscience Departments in the last research assessment (RAE 2008). The scenic Royal Holloway campus is on the outskirts of London

The studentship has a maintenance allowance of £15726 per annum for 3 years and a UK/EU tuition fee waiver . We expect candidates to have a 2.1 or first class degree (or equivalent if not a UK degree).

Apply before the 4th of March following the link ; get in touch with Tracey Jeffries ( for any application queries. If you are interested in applying please contact us informally before the deadline at or

Joint PhD position in computer vision at the University of Edinburgh and INRIA Grenoble

The LEAR research group at INRIA Grenoble and the CALVIN research group at the University of Edinburgh are looking for a joint PhD student. The candidate will be jointly supervised by Dr. Vittorio Ferrari and Dr. Cordelia Schmid. The candidate will spend time at both institutions.

The topic of the PhD is learning object classes from consumer videos. While traditional approaches learn object classes from annotated still images, in this project we want to exploit video as another source of data. A video shows a range of viewpoints, poses and lighting conditions, which are necessary for building richer models. Moreover, the temporal continuity facilitates the task of segmenting the object from its background, reducing the need for manual annotation. The PhD student will explore these aspects in order to advance the state-of-the-art in weakly supervised learning of object class models.

Starting date: as soon as possible

Applicants must have:
* Master degree in Computer Science or Mathematics
* Excellent programming skills (the project is in Matlab and C++)
* Solid mathematics foundations (especially algebra and statistics)
* Highly motivated
* Fluent in English, both written and spoken
* UK or EU nationality is mandatory
* Experience in computer vision and/or machine learning is a plus

The project will be funded by a starting and an advanced ERC grant hold by the respective supervisors. This offers a unique opportunity for the candidate to develop within a stimulating environment involving two prestigious institutions.

The School of Informatics at Edinburgh is one of the top-ranked departments of Computer Science in Europe and offers an exciting research environment. Edinburgh is a beautiful historic city with a high quality of life. For an overview of current research activities of the CALVIN team, please visit

INRIA is one of the top-ranked computer science institutes in Europe and the LEAR team at INRIA Grenoble has a track record of computer vision research (
Grenoble is locate in the French Alpes and offers excellent outdoor facilities.

This is a joint project between INRIA Grenoble and the University of Edinburgh.
The candidate will be required to spend time in both institutions.

Please send applications to the email address below, including:
* complete CV
* title and abstract of master thesis
* complete grades for all exams passed during both the bachelor and master
* the name and email address of one reference (preferably your master thesis supervisor)
* if you already have research experience, please include a publication list

Dr. Cordelia Schmid,
Dr. Vittorio Ferrari,

CFP: IJCAI Workshop on Machine Learning for Interactive Systems (abstract submission: April 13)


IJCAI Workshop on Machine Learning for Interactive Systems (MLIS’13):
Bridging the Gap between Perception, Action and Communication

August 3-4, 2013, Beijing, China

Intelligent systems or robots that interact with their environment
by perceiving, acting or communicating often face a challenge in
how to bring these different concepts together. One of the main
reasons for this challenge is the fact that the core concepts
in perception, action and communication are typically studied by
different communities: the computer vision, robotics and natural
language processing communities, among others, without much
interchange between them. As machine learning lies at the core of
these communities, it can act as a unifying factor in bringing
the communities closer together. Unifying these communities is
highly important for understanding how state-of-the-art
approaches from different disciplines can be combined
(and applied) to form generally interactive intelligent systems.

The goal of this workshop is to bring researchers from multiple
disciplines together who are in one way or another affected by
the gap between action, perception and communication that
typically exists for interactive systems or robots.
Topics of interest include, but are not limited to:

Machine Learning:
– Reinforcement Learning
– Supervised Learning
– Unsupervised Learning
– Semi-Supervised Learning
– Active Learning
– Learning from human feedback
– Learning from teaching, tutoring, instruction and demonstration
– Combinations or generalisations of the above

Interactive Systems:
– (Socially) Interactive Robotics
– Embodied Virtual Agents
– Avatars
– Multimodal systems
– Cognitive (robotics) architectures

Types of Communication:
– System interacting with a single human user
– System interacting with multiple human users
– System interacting with the environment
– System interacting with other machines

Example applications could include: (1) a robot may learn to
coordinate its speech with its actions, taking into account
visual feedback during their execution; (2) an autonomous car
may learn to coordinate its acceleration and steering behaviours
depending on observations of obstacles; (3) a team of robots
playing soccer may learn to coordinate their ball kicks depending
on the dynamic locations of their opponents; (4) a sensorimotor
system may learn to drive a wheelchair through feedback from
visual signals of the environment; (5) a mobile robot may
interactively learn from human guidance how to manipulate objects
and move through a building, based on human feedback using
language, gestures and interactive dialogue; or (6) a multimodal
smart phone can adapt its input and output modalities to
the user’s goals, workload and surroundings.

Submissions can take two forms. Long papers should not exceed
8 pages, and short (position) papers should not exceed 4 pages.
They should follow the ACM SIG proceedings format (option 1):
All submissions should be anonymised for peer-review.

Submission link:

Accepted papers will be published by ACM International Conference
Proceedings Series under ISBN 978-1-4503-2019-1. The proceedings
of MLIS’13 will be available on the ACM digital library on the day
of the workshop.

Invited Speakers:
Prof. Dr. Martin Riedmiller, University of Freiburg
Talk: “Learning Machines that Perceive, Act and Communicate”
Prof. Dr. Olivier Pietquin, Supélec, France
Title: “Inverse Reinforcement Learning for Interactive Systems”

Important Dates:
April 13, Abstract registration
April 20, Paper submission deadline
May 20, Notification of acceptance
May 30, Camera-ready deadline
August 3-4, MLIS workshop

Organising Committee:
Heriberto Cuayahuitl, Heriot-Watt University, Edinburgh, UK
Lutz Frommberger, University of Bremen, Germany
Nina Dethlefs, Heriot-Watt University, Edinburgh, UK
Martijn van Otterlo, Radboud University Nijmegen, The Netherlands

For all enquires, please mail:

PhD position at NTNU