The Sorbonne Middle for Synthetic Intelligence (SCAI) of Sorbonne College and the BnF supply a 12-month (renewable) postdoctoral contract in artificial intelligence and natural language processing.
Who are we?
Sorbonne University is a multidisciplinary research university created on January 1, 2018 by merging the schools Paris-Sorbonne and UPMC. Deploying its training to 54,000 students together with four,700 doctoral students and 10,200 overseas students, It employs 6,300 academics, teacher-researchers and researchers and four,900 library, administrative, technical, social and health employees. Its finances is 670 M€. Sorbonne College has a first-rate potential, mainly situated in the coronary heart of Paris, and extends its presence in more than twenty websites in Île-de-France and within the regions. Sorbonne University is organized into three schools: Humanities, Science & Engineering and Drugs. These schools have vital autonomy to implement the university’s technique within their very own boundaries, based mostly on a contract of goals and assets. College governance is primarily dedicated to promoting the college’s strategy, steering, creating partnerships and diversifying assets.
Presentation of the challenge
In a national and worldwide context marked by competitors around artificial intelligence, Sorbonne College has created the “Sorbonne Middle for Artificial Intelligence” (SCAI), which brings together in a single location, situated within the coronary heart of the Latin Quarter, a strategic vary of disciplines in trendy synthetic intelligence. The ambition of SCAI is to contribute significantly to the excellence of interdisciplinary research in synthetic intelligence by selling exchanges between professors, researchers, academics, students and industrialists.
The research venture described under is part of the strategic partnership between Sorbonne University and the BnF, which brings together the expertise of the MLIA group of ISIR on the BnF with a view to develop a joint analysis work with reference to recommender methods.
The Bibliothèque nationale de France (BnF) is likely one of the largest heritage libraries on the planet. Its mission is to gather, catalog, preserve, enrich and talk the nationwide documentary heritage. For many years now, BnF has been concerned in formidable digitization packages for its collections, to which we will now add the huge entry of natively digital collections. BnF is consistently enriching its digital heritage, the mass, variety and fee of progress of which require new processing and consultation tools. To allow as many people as potential to discover and applicable this heritage, BnF has been concerned in synthetic intelligence (AI) applied sciences for a number of years.
Important activities
Gallica, the digital library of the BnF, accommodates almost 10 million digitized paperwork which might be freely accessible online (18.5 million visits per yr). Nevertheless, most users do not know that Gallica accommodates not only printed documents, but in addition pictures, sound recordings, videos, and 3D objects. In satisfaction surveys, only a minority of customers think about the search engine’s solutions to be related and a majority want to be better guided of their searches. A suggestion system ought to be capable of assist customers discover their approach by way of the mass of collections and improve the visibility of the least recognized. In this undertaking, BnF is dedicated to adopting a resolutely ethical strategy. The exploitation of consumer logs must respect their privacy and assure both the relevance and transparency of the algorithms, avoiding the danger of filter bubbles. The interface design can also be at the heart of the strategy: a reliable system depends on a great consumer expertise and on the range and relevance of the proposed suggestions. Three strains of thought emerge:
- based mostly on the out there knowledge, including each consumer logs and assortment descriptions, how one can develop predictive algorithms?
- the right way to integrate variety in the suggestion algorithm while leaving the selection to the consumer to average his serendipity threshold?
- how one can build consumer belief in algorithm design and audit?
Primary missions
This challenge consists in engaged on info entry within the Gallica library, from the perspective of machine and deep learning methods. The research axes concern (1) the evaluation and indexing of textual paperwork in addition to (2) the analysis of consumer traces and (three) suggestion techniques. We're notably concerned about multimodal methods that permit contextualizing a document or a question based mostly on consumer interactions.
The successful candidate might be liable for:
- Implementing models to study the semantics of textual knowledge for the purpose of indexing them.
- Creating algorithms based mostly on representation studying methodologies to successfully mix textual content and consumer traces.
- Reporting and presenting improvement work in a clear and effective manner, both for discussion with BnF specialists and writing machine learning publications.
The printed e-book assortment will be the main focus of this system described above, but an extension to different collections with textual descriptors (particularly iconographic collections) could also be thought-about.
Schooling
A PhD degree in Pc Science or equivalent is required, as well as a robust scientific document, notably in NLP and/or Recommender Methods and/or Info Retrieval. Expertise with worldwide research tasks and purposes in SHS can be an asset.
Basic info
- Location: Pierre and Marie Curie campus of Sorbonne University and Datalab of the BnF
- Contract: 12-month fixed-term contract with the potential of an extension
- Anticipated hiring date: as quickly as potential
- Workload: full time
- Desired experience: 1 to three years
- Wage in response to expertise
Primary contacts
- Laure Soulier, MCF in pc science at Sorbonne University, MLIA workforce, ISIR.
- Emmanuelle Bermès, Scientific and Technical Assistant to the Director of Providers and Networks at BnF.
- Jean-Philippe Moreux, Scientific professional of Gallica at the BnF.
Supervision: NO
Challenge management: YES
Information and expertise
A robust background in natural language processing or text evaluation is important, and good programming expertise are required. Experience with recommender techniques is assumed. An understanding of the moral problems with such techniques can also be anticipated. Language: information of French just isn't required however is strongly most popular.
Apply
Purposes (CV + motivation + references) must be sent by e-mail to xavier.fresquet@sorbonne-universite.fr with a replica to philippe.chevallier@bnf.fr (supply crammed).
- Body: 12-month postdoctoral contract, renewable)
- Attachment: UMR 7222 ISIR
- Keywords: machine learning, explainability, databases, pc science, applied arithmetic, statistics, natural language processing, suggestion
Comments