Lambda Calculus and Differential Geometry and Big Iron

Seeking PhD students and postdocs interested in an elegant combination of functional programming and big-iron style numeric computing. (This project is, in truth, motivated by machine learning: I want to make it easy to implement complicated models, and to run them efficiently.)

Functional Programming and Automatic Differentiation

PhD Studentships
Postdoctoral Positions

We are adding exact first-class derivative calculation operators (Automatic Differentiation or AD) to the lambda calculus, and embodying the combination in a production-quality optimising compiler.
Our research prototype compiler generates object code competitive with the fastest current systems, which are based on FORTRAN. We are seeking PhD students and postdocs with interest and experience in relevant areas, such as programming language theory, numeric computing, machine learning, numeric linear algebra, differential geometry; and a burning drive to help lift big iron numeric computing out of the 1960s and into a newer higher order. Specific sub-projects
include: compiler and numeric programming environment construction; writing, simplifying, and generalising numeric and machine learning algorithms through the use of type theory and AD operators; and associated type/lambda calculus/PLT/real computation issues.

Project headquarters: Hamilton Institute, NUI Maynooth, Ireland, http://www.hamilton.ie/.

Applications and queries to:
Barak A. Pearlmutter barak+ad-fp-job@cs.nuim.ie

Action Recognition and Pose Estimation in Still Images Workshop at ECCV 2012, October 13, Firenze, Italy

http://vision.stanford.edu/apsi2012/index.html

Many human actions, such as “playing violin” and “taking a photo”, are well described by still images. Recognizing human actions and estimating human poses in still images will potentially provide useful information in image indexing and visual search, since a large proportion of available images contain people. Progress on these tasks is also beneficial to object and scene recognition, given the frequent human-object and human-scene interactions. Furthermore, as video processing algorithms often rely on some form of initialization from individual video frames, it would be interesting to have a better understanding of how, when, and to what extent static information can help recognize human actions and estimate human pose in videos. This workshop offers a great opportunity to bring together researchers and experts working on different aspects of action recognition and pose estimation to demonstrate their recent work. It provides a common playground for inspiring discussions and stimulating debates.

We invite high quality, original submissions for presentation during the workshop.
Contributions from the following areas are especially welcome:
• Human pose estimation in still images
• Human action recognition in still images • Modeling and recognition of human-object interactions • Scene context for human poses and actions • Understanding humans in videos or depth images • Novel datasets of human poses or actions • Actions and human pose research in cognitive psychology / human perception

Papers must be in PDF format and must not exceed 10 pages (ECCV format). All submissions are subject to a double-blind review process by the program committee.

• Deadline for submission of papers: July 3rd, 2012 • Notification of acceptance: August 2nd, 2012 • Camera-ready submission: August 8th, 2012 • Workshop date: October 13rd, 2012

The organizers can be contacted through apsi2012.ws@gmail.com

Open PhD position in machine learning and signal processing at Qarma, LIF, Marseille

We are looking for candidates for a three-year funded PhD entitled “Acceleration of greedy algorithms for signal modeling – Theory and applications” at Qarma, LIF, Marseille. See offer below, also available at http://www.lif.univ-mrs.fr/~vemiya/research/PhD_offer_Greta_Qarma.pdf .

Acceleration of greedy algorithms for signal modeling – Theory and applications

PhD position

Supervisors: Valentin Emiya and Liva Ralaivola, Laboratoire d’Informatique Fondamentale de Marseille, France

This PhD position is funded for three years.

PhD description

Keywords. Algorithms, machine learning, signal processing, optimization, sparsity.
Abstract. This PhD project addresses theoretical and applicative aspects of emerging sci- entific topics at the interface between machine learning and signal processing. The goal is to design new greedy algorithms, focusing on their speed.

Scientific context. Sparse approaches can model a wide range of signals – images, sounds,
biomedical signals, and so on – and their intrinsic structures. They benefit from a solid theory and efficient algorithms. These modern techniques are used in a growing number of applications such as signal compression and coding, source separation, inpainting or pattern recognition.
Goal. Today’s algorithms are limited by their computational complexity. As a result, they are slow and cannot handle large volumes of data. In practice, this is a major limitation since applications are more and more demanding e.g. in terms of real time constraints or fine 3D discretization. Recent theoretical results have opened new directions for the design of fast algo- rithms. In the case of greedy algorithms such as Matching Pursuit, preliminary theoretical and experimental results have been obtained at Laboratoire d’Informatique Fondamentale de Mar- seille (LIF). The goal of this PhD project is to develop this topic by designing new algorithms and validating them with theoretical and practical considerations. Several applications will be addressed such as audio inpainting and underwater acoustic imaging.

Tentative workflow. First, the PhD candidate will get familiar with sparse models and algorithms, and work on algorithmic ideas emerging from the preliminary results obtained at LIF. The research will be pursued in both the theoretical and the applicative direction, the focus being more on one or the other, depending on the candidate’s profile.

Collaborative context. The PhD project is at the core of a French ANR (national fund) project called GRETA (Greediness – theory and algorithms) initiated in 2012 by LIF, LATP and INSA Rouen. The PhD candidate will work within an active research group and will have the opportunity to travel inside and outside France, for conferences and collaborations.
Supervisors: Valentin Emiya and Liva Ralaivola, Qarma team, LIF.

Team and scientific environment
The PhD candidate will join the Qarma team at Laboratoire d’Informatique Fondamentale de
Marseille, which is a joint research lab between CNRS and Aix-Marseille University. The Qarma team (about twelve researchers) dedicates its activities to statistical machine learning theory and applications to signal processing and multimedia indexing.

The PhD candidate will also work with Sandrine Anthoine, a researcher at the neighboring mathematics lab. LATP. Other scientific exchanges will be possible locally and nationally (with the partners of the GRETA project, Rémi Gribonval at INRIA Rennes, Laurent Daudet at Paris 7 University, Jacques Marchal at Paris 6 University).
Required background

The applicant must have at least one of the following profiles: maths, signal processing, machine learning, computer science.

Salary
The candidate will be hired for three years with a standard French PhD contract (contrat doc- toral). The net salary is 1374.21 euros per month including health insurance. Teaching or consulting opportunities may be added to the contract if the candidate wishes.

Application
Applicants should contact Valentin Emiya and Liva Ralaivola (firstname.lastname@lif.univ-mrs. fr) as soon as possible. Additional information about the PhD project will be available at this stage.
The application must be sent by email to Valentin Emiya and Liva Ralaivola according to the schedule below. It must include

• a cover letter explaining the applicant’s motivations,
• a resume,
• a transcript of the Master degree marks (or equivalent), • optional: recommendation letters or referees contacts.

The selected applicants will be interviewed, either in the lab or by phone.
Schedule (subject to changes):

• June 30, 2012: deadline for sending the application.
• July 10, 2012: end of the interviews and announcement of the results. • October 1st, 2012: beginning of the contract.

Contact
Valentin Emiya and Liva Ralaivola, firstname.lastname@lif.univ-mrs.fr.

Post-doctoral Researcher in the area of data mining (Lugano, Switzerland)

A position for a post-doctoral data mining researcher is available at IDSIA (Lugano, Switzerland, www.idsia.ch).

**Duties
-Research and development of original data mining methods for classification, regression and clustering.
-Implementation of data mining algorithms within an existing Business Intelligence software program.

**Requirements
– Master’s Degree in computer science, statistics, physics, engineering or other highly quantitative areas.
-PhD in data mining related topics, or in statistical data modeling, preferably involving Bayesian networks.
-Strong publication record in the fields of data mining, machine learning or computational statistics.
-Strong commitment to research and scientific publication, together with the willingness to cooperate with a leading business intelligence company.
-Strong background in programming languages such as C++, C# or Java.
-Ability to produce high-quality data mining implementations, to efficiently analyze very large databases.
-Excellent oral and written English skills.

**We offer
-A two-year position with an employment contract of 100% and with the possibility of contract extension.
-International working environment.
-Collaboration with expert data mining researchers and with leading Business Intelligence companies.
-Participation in prominent international conferences, with the possibility for funded travel.
-Attractive salary, in line with Swiss employment standards.

**Applications
Applicants should submit the following documents, written in English:
– curriculum vitae
– list of exams and grades obtained in the course of studies conducted at Bachelor’s and Master’s level;
– names of three references (with e-mail addresses); -letter of motivation: brief statement on how the applicant’s research interests correspond with the above mentioned topics (1 – 2 pages); -list of publications, and, if possible, link to the thesis.

Completed applications should be submitted by the 5th July 2012 using the online form found at the following address:
www.supsi.ch/go/bando_data_mining_idsia

For further information please refer to the official page for this
position:
http://www.supsi.ch/home/supsi/lavora-con-noi/2012-07-05.html

or contact
Giorgio Corani
giorgio.corani@supsi.ch

*** MacBrain Symposium ***

http://macbrain2012.cs.ru.nl/
6 September 2012; Nijmegen, The Netherlands

A one day international symposium on the integrative analysis of brain diseases: brain banks, machine learning, clinical and pathological studies.

Speakers

* Jean-Paul Vonsattel – Director of the New York Brain Bank at Columbia University, USA
* Daniel Lightfoot – Director of the Autism Tissue Program, USA
* Sophia Ananiadou – Director of the National Centre for Text Mining, UK
* Yves Moreau – Program director of POC Bioinformatics Department of Bio-Engineering sciences, KU Leuven, Belgium
* Amanda Kiliaan – Department of Anatomy/Cognitive Neurosciences, Radboud University, NL
* John C. van Swieten – Erasmus MC, NL
* Inge Huitinga – Netherlands Brain Bank, NL
* Elena Marchiori – Radboud University, NL

Overview

The goal of this symposium is to provide researchers and practitioners an overview of the cutting edge research on core aspects of integrative research of brain diseases, combining data management, machine learning and clinical and pathological studies.
The lectures from worldwide domain experts will describe state-of-the-art research and discuss innovative directions aiming at optimizing knowledge discovery processes, from data collection in brain banks through machine learning to integrative clinical and pathological studies.

The symposium will provide a multi-disciplinary forum for discussing issues and
achievements in research on brain diseases. It will stimulate and strengthen synergy and collaboration among researchers from different disciplines, such as brain bankers, data managers, IT specialists, pathologists and clinicians, in order to improve the efficiency and effectiveness the integrative study of brain diseases.

Registration

Registration is on a first-come first-serve basis, so register as soon as possible at
http://macbrain2012.cs.ru.nl/registration. The registration fee for the MacBrain 2012 symposium is €50. It includes coffee/drinks and lunch.

Call for Papers: ECMLPKDD Workshop on Mining and exploiting interpretable local patterns

September 28, 2012, Bristol/UK
Submission deadline: June 29, 2012

Workshop homepage:
http://www.iais.fraunhofer.de/interpretable-patterns-workshop.html

———————————————————————–
Topics of interest:

* Actionable patterns
* Applications of local pattern mining, e.g. in clinico-genomic,
fraud detection or marketing settings
* Interactive data-mining
* Interpretable models
* Measures and optimization of interestingness for rules and models
* Pattern ordering and pattern set selection
* Scalability of local pattern mining
* Subgroup discovery

———————————————————————–
Submission details:

The papers must be written in English and formatted according to the Springer LNCS guidelines. Authors instructions and style files can be downloaded at: http://www.springer.de/comp/lncs/authors.html. The maximum length of the papers is 16 pages.

The complete submission process will be managed via Easy Chair:
https://www.easychair.org/conferences/?conf=ipat2012.

———————————————————————–
Detailed description:

Local patterns, like itemsets, correlations, contrast sets or subgroups, are valuable nuggets for a variaty of applications. Among others, they can been used for classification, regression or outlier detection tasks. One particular characteristic which makes them stand out from other machine learning tools, however, is that (most) local patterns can directly be read and interpreted by end users lacking a profound machine-learning background. This descriptive nature of local patterns makes them useful as a source of information for decision making. For example, in the analysis of clinical data, understandable models can help the clinician in understanding his data and thus making an informed decision about patient treatment. In addition, understandable knowledge can help domain experts to discuss the analysis results and collaboratively find a good, interesting solution in data-intensive settings to help guide the learner when complex background knowledge prevents the system from finding a good model without further input.

In this workshop, we wish to investigate typical use cases and key requirements for the successfull usage of local pattern mining in applications where next to the statistical performance of models, the understandability and interestingness of the models is the key success factor. Here, we are particularly interested in settings where the data to be mined is large and complex, preventing investigations of the data without (semi-)automatic analysis tools. Key questions to be investigated in the workshop are: which pattern language is adequate both for the representation of local phenomena and for the interpretation by the user. How can the raw set of local patterns be reduced to a representative and manageable subset?
What conditions must be satisfied for a pattern to be actionable?
How can feedback about the understandability and interestinness of partial models be given back to the system and how can the search be controlled? What is the best way to deal with the (typically exponential) size of the pattern space? How to design scalable algorithms? Beside papers focusing on such questions, we welcome case studies of successful descriptive pattern mining applications. Papers combining applications and theoretical contributions are particularly welcome.

PRELIMINARY CALL FOR PAPERS – IbPRIA 2013 – 6th Iberian Conference on Pattern Recognition and Image Analysis

Madeira, Portugal
June 5-7, 2013
http://www.ibpria.org/2013

The Iberian Conference on Pattern Recognition and Image Analysis (IbPRIA) is an international event co-organised every two years, by the Portuguese and Spanish Associations for Pattern Recognition.

IbPRIA is a single track conference that includes tutorials, invited speakers as well as oral and poster presentations. The conference is intended to act as a forum for research groups, engineers and practitioners to present recent results, algorithmic improvements and promising future directions in pattern recognition and image analysis.

SCOPE
The conference is looking for new theoretical results, techniques and main applications on any aspect of pattern recognition and image analysis, including but not restricted to the following topics:

Pattern Recognition
Image Analysis
Computer Vision
Multimedia Systems
Statistical and Structural Pattern Recognition
Machine Learning and Data Mining
Bioinformatics
Image Coding and Processing
Shape and Texture Analysis
Information Systems
Biometric Technologies
Speech Recognition
Document Processing
Character and Text Recognition
Robotics
Remote Sensing
Industrial Applications of Pattern Recognition
Special Hardware Architectures

PAPER SUBMISSION
Papers should describe original and unpublished work on the topics of the conference. Prospective authors should prepare a full paper, written in english, not exceeding 8 pages and must submit it electronically. Further information can be found on the conference website: http://www.ibpria.org/2013
Each paper will be blind-reviewed by at least two reviewers and will be accepted based on its originality, signficance and clarity. All accepted papers will appear in the conference proceedings, scheduled to be published in Springer LNCS series. A copy of the proceedings will be distributed to all participants at the Conference.
Moreover, the IbPRIA 2013 conference provides additional opportunities for journal publicaction and best paper awards. Contacts are being established with an indexed journal for a special issue.
Submission implies that at least one of the authors has to register and to present the communication at the conference if the paper is accepted.

IMPORTANT DATES (TENTATIVE)

Submission of papers: November 19, 2012
Notification of acceptance: January 19, 2013
Camera-Ready: February 2, 2013
For more information please visit http://www.ibpria.org/2013

Open PhD position at the ERIC Lab in Lyon, France

A PhD position is now open at the ERIC Lab for the development of “New graphical models for temporal clustering”. All applications are welcome, with an expected start date about the 1st of September or the 1st of October 2012. Detailed description of the PhD is available here:

http://eric.univ-lyon2.fr/~jvelcin/public/misc/phd-Imagiweb-en.pdf (english version)
http://eric.univ-lyon2.fr/~jvelcin/public/misc/these-Imagiweb-fr.pdf (french version)

Deadline for application: 8th of July, 2012.

PhD scholarship at INRIA – LEAR research group in Grenoble, http://lear.inrialpes.fr

Topic: The goal of this PhD is to better understand videos by exploiting associated textual data. For still images some success has been achieved in learning correspondences between objects and textual keywords [1]. In video, it has been demonstrated that transcripts aligned with the video can be a very useful source of weak supervision for learning the appearance of characters [2] and human actions [3].
Existing work does not attempt to use text as a form of supervision for learning spatio-temporal constraints between scenes, humans, objects and their interactions in video. In addition, the text is typically considered as a supervisory signal for visual learning and the opposite direction, where visual information would help disambiguate text interpretation, is not considered. In this PhD, we propose to go beyond the state-of-the-art and turn textual annotations into a more complete and accurate supervisory signal for the different stages of the scene/object/human action interpretation process. In particular, we want to develop spatio-temporal correspondences between videos and the available text annotations, and exploit these correspondences as constraints for learning actions in videos.

Your profile:

* Master degree (preferably in Computer Science or Applied Mathematics; Electrical Engineering will also be considered)
* Solid programming skills; the project involves programming in C
* Solid mathematics knowledge (especially linear algebra and
statistics)
* Creative and highly motivated
* Fluent in English, both written and spoken
* Prior knowledge in the areas of computer vision, machine learning or data mining is a plus (ideally a Master thesis in a related field)

Duration: 3-4 years

Start date: September 2012.

Location: INRIA Grenoble, France. Grenoble lies in the French Alpes and offers ideal conditions for skiing, hiking, climbing etc.

Contact: Cordelia Schmid, schmid@inria.fr

Please send applications via email, including:

* a complete CV
* graduation marks
* topic of your Master thesis
* the name and email address of two references (including your Master thesis supervisor)

Literature:

[1] M. Guillaumin, T. Mensink, J. Verbeek and C. Schmid.
International Journal of Computer Vision, 2012.

[2] M. Everingham, J. Sivic, and A. Zisserman. Taking the bite out of
automatic naming of characters in TV video. Image and Vision
Computing, 2009.

[3] O. Duchenne, I. Laptev, J. Sivic, F. Bach, and J. Ponce. Automatic
annotation of human actions in video. In International Conference on
Computer Vision, 2009.

CVMP2012 Full Paper deadline EXTENDED to 26th June.

CVMP2012 Full Paper deadline EXTENDED to 26th June. Don’t forget the Google £2k Best student paper prize this year.

If you have been doing innovative work then we’d love to hear about it. High-quality papers are invited which present novel research related to any aspect of media production. Full length submitted papers will be subject to peer review. The papers will be published in cooperation with Eurographics and ACM (pending). Selected papers will be invited (and fast-tracked) to submit extended versions to the IEEE Transactions on Multimedia. All details on the CVMP website http://www.cvmp-conference.org/

CVMP 2012 is the 9th annual industry-academic conference series on media production for film, broadcast and games. Visual media production brings together expertise in video processing, computer vision, computer graphics, animation and physical simulation. CVMP provides a forum for presentation of the latest research advances combined with key-note and invited talks on state-of-the-art industry practice in content production and post-production. CVMP is rapidly building a reputation as a prime venue for researchers to meet with image practitioners in the film and television industry. For instance, in 2011 48% of the audience were from the film and television post-production industry including representation from high profile post-houses such as Moving Picture Company, Double Negative, and so forth. Furthermore, CVMP hosts keynotes from distinguished researchers and professionals from industry (to name have few from previous years: Jeremy Doig / Google, Roberto Cipolla / Cambridge University, Sylvain Paris / Adobe).