CONTEXT - The generation in quasi-continuous flow of an increasingly massive quantity of data prohibits in many fields to carry out the labeling phase of supervised learning.
In fact, this task can no longer be carried out by experts because too many tedious and time-consuming. Moreover, it assumes that the experts already have an a priori definition of the classes (nomenclature, ontology, etc.) that may be of interest to them. However, this is not always available or is only partial.
the ANR HERELLES(+) project aims to meet this need by proposing an innovative method of collaborative learning interactive multiparadigm, allowing to combine supervised and unsupervised methods while allowing interaction with the expert.
This project is based on two complementary aspects, the use of exchanges of information between methods inspired by existing concepts, boosting [3], co-learning [4] for the supervised and collaborative [2] for the unsupervised, as well as the addition constraints in this process [1].
SUBJECT DESCRIPTION
The recruited person will have to propose and define original mechanisms allowing supervised and unsupervised methods
supervised to collaborate effectively to arrive at classification consensus. Modalities for exchanging information
between them should be specified. It must also define an interaction protocol between the user and the methods learning through the use of constraints.
Finally, it will have to concretely implement the approaches proposed to allow the testing and validation of these.
COLLABORATION AND SUPERVISION
The recruited person will be co-directed by Antoine Cornuéjols (AgroParisTech - 50%), specialist in supervised learning collaborative and Pierre Gançarski (ICube - 50%), specialist in collaborative clustering.
She will actively collaborate with the SDC team of ICube in Strasbourg and more particularly with Antoine Saget doctoral student (1st year) working on the same subject, as well as Baptiste Lafabregue, co-author, with P. Gançarski, of several articles on constrained clustering of time series [1].
She may therefore have to travel there frequently (at the expense of the laboratory).
GENERAL INFORMATIONS
Location: Saclay (AgroParisTech Campus, 22 place de l'Agronomie, 91120 Palaiseau)
Duration: One year (renewable once)
Salary: from 2500€/month to 2700€/month (gross) depending on experience.
Contact: Antoine Cornuéjols [email protected] and Pierre Gançarski, [email protected]
DESIRED PROFILE
- PhD in computer science and specialized in machine learning/data mining.
- Solid knowledge in Data Science and more particularly on standard methods of classification and clustering. A first experience on the use of collaborative/set models or integration of constraints would be one more.
- Good verbal (English or French) and written (English) communication skills.
- Interpersonal skills and the ability to work individually or as part of a project team.
TO APPLY
Interested persons must submit (by email to [email protected]) their curriculum vitae, a list of their publications, a cover letter and the contact details of three references. Applications will be admitted until the position is filled. The position will start as soon as possible.
Attached is an offer for a 24-month research engineer position.
Project : SmartFCA (ANR), 2022-2026
Laboratory : IRISA, Rennes
Team : LACODAM
Duration : 24 months, starting between February and April 2023
Contact : [email protected] , [email protected]
Keywords : data mining, Formal Concept Analysis (FCA), graphs, interoperability, linguistic data
Background : Formal Concept Analysis (FCA) [1] is a knowledge discovery method. It is used in data analysis, data mining, classification or information retrieval tasks; and applied in various fields such as life sciences, human sciences or linguistics. Multiple extensions of FCA have been proposed by different teams to process complex data such as sequences, trajectories, trees or graphs [2]. Beyond the theoretical and practical locks, there is a problem of interoperability between these different extensions, which hinders their adoption and their composition in workflows.
An important goal of the SmartFCA project is to make these FCA extensions interoperable by encapsulating them in software components with conceptually and technologically compatible interfaces. It is also a question of implementing a platform allowing the construction of workflows from components. The IRISA/Rennes partner is responsible for the Graph-FCA component [3], an extension of FCA to relational data and graphs. We work closely with the partner ICube/Strasbourg who is responsible for the component for another extension of FCA to relational data, RCA (Relational Concept Analysis) [4].
Another objective of the project is to develop use cases in various fields, for their intrinsic interest and to evaluate the developed platform. IRISA/Rennes will develop use cases on the linguistic data of poorly endowed languages (Breton [5,6] and Georgian [7] in particular).
Subject : After a familiarization phase with Graph-FCA and its current implementation, as well as RCA, it will be a question of collaborating with ICube/Strasbourg to design a compatible interface between the two FCA extensions (input/output modeling, option sets ). It will then be a question of encapsulating the existing implementation of Graph-FCA in a RESTful API, in accordance with the standards established within the framework of the project. The candidate is expected to collaborate with the other partners of the project in the establishment of these standards, and to be proactive. It will also be necessary to develop test and demo interfaces of the Graph-FCA component so as not to depend on the platform, which will only be completed towards the end of the project.
The candidate will also have to provide technical support and be proactive in linguistic use cases (no knowledge of linguistics is required). This includes helping to prepare the data, applying the Graph-FCA component and the other components developed in the project and highlighting the results, that is, the knowledge extracted from the data.
Candidate profile : We are looking for a candidate motivated by research & development experience as part of an academic research project. The required training is a doctorate or a master's degree in computer science.
Expertise required for the position:
web programming, especially backend and Node.js: design, development, configuration and documentation
data models, including relational and graphs
development tools and methods
teamwork
writing technical reports and oral presentations
Desired knowledge or experience:
knowledge extraction (data mining, data mining, classification)
Caml programming or other functional language (Haskell, Scala, …)
Expected qualities: autonomy, rigor, ability to collaborate face-to-face and remotely with several teams, strength of propo
We have two open positions in deep learning at the University of Toronto. Both positions are for a tenure-track assistant professor.
I would appreciate if you could circulate the following links to relevant candidates and mailing lists.
https://jobs.utoronto.ca/job/Toronto-Assistant-Professor-Deep-Learning-ON/565386217/
(apply by January 9, 2023)
https://academicjobsonline.org/ajo/jobs/22985 (apply by December 30, 2022)
Candidates will also be reviewed by the Vector Institute for Artificial Intelligence for consideration of appointment as a Faculty Member or Affiliate Faculty Member, and may be nominated for a Canada CIFAR AI Chair through the Vector Institute.
Project: IA2: Explainable AI as an interface for algorithmic auditing (Leonardo Grant Fundación BBVA)
Context Marie Curie Project: RRR-XAI: Right for the Right Reason eXplainable Artificial Intelligence
https://cordis.europa.eu/project/id/101059332
Contract duration: 6-36 months
Directors: Natalia Díaz Rodríguez & Francisco Herrera, UGR.
Start: Immediate
Project summary: The objective of this project is to try to overcome some deficiencies of Artificial Intelligence based on deep neural networks, a form of machine learning that allows machines to learn from experience. First of all, these networks are considered black box models, that is, with opaque algorithms, which do not allow their results to be interpreted and diagnosed. Second, they suffer from bias and can misinterpret the data, mistaking mere correlations for causal relationships. Díaz's project will study two practical use cases – the prediction of COVID-19 in chest X-rays, and the detection of weapons in crowds from images – to try to develop deep neural networks that are more transparent and that provide a consistent explanation of the results it achieves.
More info: https://www.redleonardo.es/beneficiario/natalia-diaz-rodriguez/ Contact: CV+grades+github to [email protected]
Apply formally: until 22nd December at: Project Reference 251, Pag. 52: https://investigacion.ugr.es/recursos-humanos/personal/contratos
In the context of the EOSC related Horizon Europe FAIR-IMPACT project (https://fair-impact.eu) INRAE’s MISTEA research unit is hiring a project manager in Open Science, Metadata and Ontologies.
Please see the offer here: https://jobs.inrae.fr/en/ot-16522
Thanks for transferring the info.
Tweet is here: https://twitter.com/jonquet_lirmm/status/1600904576403738635
Combinatorial optimization consists in finding the best solution among a finite, but very large, set of possible configurations. It is a branch of discrete mathematics and computer science that has applications in many fields, such as economics, logistics, energy or health.
For the last ten years, the LERIA1 (Laboratoire d’Etude et de Recherche en Informatique d’Angers) has been engaged in new exploratory work that breaks with conventional approaches to solving combinatorial optimization problems.
A new generation of methods using machine learning techniques has been proposed. The general objective of these methods is to make the solving algorithms more autonomous, and to make them adapt in real time to the problem at hand.
These works have given promising results and have already led to the publication of reference articles on dynamic island models [1], on pattern extraction in solutions [2] and the development of reinforcement learning methods for graph coloring problems [3, 4].
This project aims to continue in this direction and to develop this time resolution methods for combinatorial problems based on deep neural networks (deep learning), in the manner of the spectacular advances that have taken place in recent years for combinatorial games such as the game of go or chess with the AlphaZero algorithm [5].
Very recent works have already proposed to apply such methods of deep learning to solve combinatorial optimization problems, such as travelling salesman problems or graph coloring problems [6, 7]. Nevertheless, these works exploit little or no specific knowledge of the problem, which greatly limits the interest and performance of these approaches.
In fact, the results obtained by this type of methods are for the moment very far from the state of the art results mainly obtained by memetic algorithms [8] and simulated annealing algorithms [9].
The objective of this postdoc will be to develop new hybrid approaches that combine deep neural networks with the best tools of ”classical” metaheuristics in order to solve combinatorial optimization problems that still resist the best state-of-the-art methods.
A first work has recently been published by our team on this subject [10].
The duration of the contract will be 18 months, the start date is expected between January and July 2023 and the salary will be approximately 2150 euros net.
Profile We are looking for PhD with the following requirements: • Solid background in metaheurstics and/or deep learning.
• Strong development skills in C++ and/or Pytorch.
Documentation - Candidates who are interested in applying, please send the following documentation to olivier.goudet@univ-angers. fr
• Short CV (3 pages maximum);
• The list of at least three publications;
• One contact willing to support your application.
Contexte: Blind and semi-blind unmixing problems are ubiquitous in a very wide range of scientific domains from remote sensing to astrophysics. In these domains, the fast development of high resolution/high sensitivity multispectral sensors mandates the development of dedicated analysis tools to extract relevant and interpretable information. With the fastly increasing size of the multi/hyperspectral to process, one of the key challenges to overcome is the development of fast and yet efficient solvers for multispectral data unmixing. The goal of the project is to explore and develop novel machine learning based approaches based on algorithm unrolling to tackle these problems.
Activity: The postdoc activity will focus on the development of dedicated algorithm unrolling models to tackle blind/semi-blind multispectral unmixing problems. This work will particularly focus on investigated approaches with low level of supervision, which is fundamental in physical applications. The developed algorithms will be tested and validated on X-ray multispectral images in astrophysics in preparation for the Xrism and Athena space mission.
Skills: PhD level in computer science or signal/image processing. Excellent coding skills in python, with a good knowledge of pytorch.
Team: The candidate will be part of a collaboration between the computer science lab at CEA/IRFU (CEA Paris-Saclay) and the Image team at Telecom ParisTech.
General information
Funding: two years.
Gross salary starting from 30kEuros/year depending on the experience of the candidate.
Position available now.
Position open to non-EU citizens.
More info: https://www.dropbox.com/s/shyplokthuxbedr/Postdoc_HybridMethods.pdf?dl=0
This Post-doc will be done in the ConfianceAI framework (Confiance.ai). Confiance.ai is the technological pillar of the Grand Défi “Securing, certifying and enhancing the reliability of systems based on artificial intelligence” launched by the French Innovation Council.
It is the largest technological research programme in the #AIforHumanity plan, which is designed to make France one of the leading countries in artificial intelligence (AI).
This post-doc aims to design models to estimate and predict CNN performances without requiring their deployment on the real platform. These models will save time since measurement campaigns will be needed.
Furthermore, it will be possible to test CNN architectures and measure their performances, in terms of execution time, power consumption, memory occupation, etc, before the platforms being available in the market.
Thus, measurements such as execution time, or energy consumption can be estimated on an architecture not available from the manufacturer if we know its most "impacting" attributes on the Hardware architecture(s).
This solution makes it possible to quickly explore a large space of configurations (CNN architectures, hardware architectures).
It also makes it possible to avoid developing and optimizing models that will turn out ineffective on the target platform. In this postdoc we will develop new models for CNN performance prediction that respect the rank of the models (Pareto rank-preserving surrogate models).
Hardware-aware Neural Architecture Search (HW-NAS) has recently gained steam by automating the design of efficient DL models.
However, such algorithms require excessive computational resources and thousands of GPU days are needed to evaluate and explore CNN search space.
In this Postdoc we will explore the effectiveness of multi-objective Surrogate Performance Prediction Models in HW-NAS.
Tasks that have to be developed in this postdoc are :
- Determination of the CNN search space
- Determination of the main characteristics for two of Hardware platforms. This may concern
Qualcomm Hexagon DSPs and edge Nvidia GPUs.
- Development and testing of multi-objective surrogate performance prediction models for these 2 types of HW platforms.
Education: A Ph.D in computer/electrical science/engineering is required
Salary: 2300 euros/month net (3000 euros/month gross)
Deadline for application: 30/09/2022
Duration : 1 year (1 year extension possible)
Employer : IRT Systemx
Address: 2 Bd Thomas Gobert, 91120 Palaiseau
Start date: Preferably between Sept 1st 2022 and Nov 1th 2022.
An application prepared in English or French should contain:
1. CV with the list of publications.
2. Contact information for 2-3 reference persons.
3. Your most relevant conference or journal publications, in full-text.
For further information contact: - Prof. Smail Niar ([email protected]) LAMIH/CNRS, INSA Hauts-de-France and CNRS, France
- Prof Elghazali Talbi ([email protected]), INRIA and University of Lille, France
We are seeking a postdoc candidate to develop machine learning approaches for understanding regulatory genomics, i.e. to understand how regulatory DNA sequence operates to control gene expression.
This is a fundamental question in cellular biology, with numerous applications.
In medicine for example, this is an important issue to understand the molecular basis of cancers, which are pathologies invariably associated with a deregulation of gene expression.
A detailed description of the project can be found here: https://emploi.cnrs.fr/Offres/CDD/UMR5535-CHALEC-002/Default.aspx
Being familiar with genomics is an advantage but this knowledge can be acquired in the course of the project with interactions of members of the consortium and/or with dedicated theoretical courses and workshops.
Contacts : Laurent Bréhélin (LIRMM) [email protected]
Charles Lecellier (IGMM) [email protected]
IRISA - CNRS (Rennes, France) offers a post-doc position (18 months) in TAL. The recruited person will work on the hybridization between symbolic learning and deep learning for the text, and on the robustness of the classifiers within the framework of an ASTRID/AID project on misinformation.
Details and application: https://emploi.cnrs.fr/Offres/CDD/UMR6074-VINCLA-008/Default.aspx
The admissibility of the application is potentially subject to review by the Defense Innovation Agency.
Contact: [email protected]
IRISA-CNRS (Rennes, France) offers a post-doc position (18 month contract) on NLP. The recruited researcher will work on symbolic/deep learning hybridization for text and on classifier robustness in the framework of an ASTRID/AID funded project about disinformation.
Details and application: https://emploi.cnrs.fr/Offres/CDD/UMR6074-VINCLA-008/Default.aspx?lang=EN
The application may be reviewed by the Agence Innovation Defense.
Contact: [email protected]
A postdoc position (up to 18 months) is currently open. The position is financed by the JAPETUS project, which aims at operating a constellation of earth observation nano-satellites allowing high-frequency fly-bys and quick reactivity. The first phase of the project includes the launch of a prototype in 2025.
The research will take place at LAAS-CNRS in Toulouse and is concerned in particular by the design of efficient methods to plan the acquisitions and downloads of the JAPETUS constellation. The JAPETUS constellation will be composed of 20 satellites on inclined orbits, allowing for very frequent fly-bys, and it is capable of handling urgent preemptive requests thanks to inter-satellite communication. These features make the planning problem extremely challenging.
We seek motivated candidates with a strong background in computer science, with excellent programming skills and some previous knowledge and experience in solving combinatorial optimisation problems.
The deadline to apply is January 15 2023 and the contract may start in March 2023.
Application link: https://emploi.cnrs.fr/Offres/CDD/UPR8001-EMMHEB-002/Default.aspx
Contact: Emmanuel Hebrard [email protected]
A more detailed description of the PostDoc offer can be found here: https://homepages.laas.fr/ehebrard/doc/Post_doc_JAPETUS.pdf