Blog

Projets 2023 pour preparer OLKi2

Wednesday, Sep 21, 2022 by Christophe Cerisara OLKi2

Projets LUE 2023 pour OLKi2 texte de l’appel sur la BUL Liste des projets soutenus par OLKi (L’ordre des projets est totalement aléatoire et ne doit en aucun cas être considéré comme lié d’une manière ou d’une autre à un ordre préférentiel) Karen Fort et Anna Zielinska: éthique de l’IA (AIRed) Yannick Parmentier et Mathieu Constant: grammatiseur (EsPaR) Julien Falgas et Christophe Cerisara: modèles de langage et Needle (SincerePLM) Irina Illina et Céline Ségur: hate speech (Theodesm) Maxime Amblard et Jean-Sébastien Rey: Language et bible (Ben Sira Transformation Overtime) Alain Polguère et Francesca Ingrosso: vocabulaire de la chimie (T-to-T-Chem) Argumentaire Tous ces projets visent à analyser le discours et le dialogue en contexte sociétal.

Post-doctoral position available

Tuesday, Jul 7, 2020 by Karen Fort ethics

JOB DESCRIPTION This post-doc contract is funded by the EU Horizon2020 Artificial Intelligence for improved PROduction efFICIEncy, quality and maiNTenance (AI-PROFICIENT) project. It aims at investigating ethics by design in the context of the industry of the future and more particularly the contribution of digitalization and Artificial Intelligence to predictive maintenance issues and production optimisation including human-system interaction. It is therefore situated at the intersection between machine learning ethics, human-machines and human-systems interactions and data management issues.

OLKi platform: quick install guide

Thursday, Jun 25, 2020 by Christophe Cerisara AI

Install the OLKi platform ** A more recent version of this post is available here !** A common situation is the following, where we assume that you have: An ubuntu computer with apache + docker A personal domain name, say “olki.toto.fr”, which will be your entry point to your instance Then the easiest way to install the platform is to use the latest docker image. After cloning the OLKi repository, there are 3 configuration files that are important, as explained in the official documentation:

TALN workshops

Tuesday, Jan 21, 2020 by Christophe Cerisara AI

One workshop and one tutorial linked with OLKi have been accepted at the TALN conference: ETeRNAL: about ethics and NLP CLARIN-FR: about language resources and platforms Follow us on Mastodon ( @admin@olkichat.duckdns.org ) or have a look at the Mastodon stream on the landing page of OLKI to get more news in the next few weeks about these workshops !

Claire Gardent is granted an AI Chair

Saturday, Dec 14, 2019 by Christophe Cerisara AI

Claire Gardent has been granted an AI Chair Claire Gardent, CNRS researcher in the Synalp research team, is already internationally recognized for her contributions in Natural Language Processing and Artificial Intelligence. She has been granted in December 2019 a national “AI Chair”: only 3 chairs have been granted in the Région Grand Est. This is a great news for the OLKi project, and the development of Artificial Intelligence in the region !

A deep learning dictionary

Friday, Dec 13, 2019 by Christophe Cerisara AI

A deep learning dictionary Here is the first version of a succint, but a bit off-focus, deep learning dictionary.

Machine Learning on copyrighted data

Tuesday, Nov 19, 2019 by Christophe Cerisara AI, deep learning

Training algorithms on copyrighted data is not illegal, according to the United States Supreme Court According to a recent Supreme Court decision in the USA, it is allowed to train machine learning models on copyrighted data: https://towardsdatascience.com/the-most-important-supreme-court-decision-for-data-science-and-machine-learning-44cfc1c1bcaf This decision is very impacting in the current debate about data ownership and privacy, but it certainly should not be interpreted in an oversimplistic way, as there is always the problem of re-generating the source data from a number of recent deep learning models, which is likely to moderate these conclusions in many practical cases…

Federated Deep Learning

Wednesday, Oct 23, 2019 by Christophe Cerisara AI, deep learning

Challenge of training deep learning models Training deep learning models has always been a challenge. Not only because of the large quantity of data to train on, or because of the large number of model parameters to optimize, but also because of the fact that training a deep learning model involves finding its high through a huge variety of related paths, which involves testing many different combinations of hyper-parmeters and model topologies, without any certainty but just using basic intuition and human knowledge and expertise.

What is OLKi ?

Tuesday, May 21, 2019 by Christophe Cerisara project

OLKi is a research project. An outcome of the project will be a platform, which is: For everybody, to share resources, any kinds of resources (files, programs, datasets…), and to communicate. You stay in control of your resource, and you benefit from the federation. You don’t need to install anything and may choose to be hosted, if you prefer. It’s not only for everybody, it’s also for scientists, professionals and amateurs; it’s

Summer school in NLP

Monday, Apr 29, 2019 by Christophe Cerisara

We organize a summer school about python programming for Natural Language Processing, in Nancy, in the last week of August. Please go there for more details !

Ethique des usages des données

Sunday, Apr 21, 2019 by Christophe Cerisara AI, society

Réunion ouverte sur l’éthique Le débat est ouvert vendredi 24 mai, de 16h à 18h, au LORIA, en salle B013 ! Si cela vous intéresse, envoyez-moi un email ou prévenez-moi sur Mastodon. Problématique Les données sont présentes partout et les mouvements contestataires isolés (“aucun portable chez moi !”) ne résoudront rien; il faut changer de modèle. De nombreuses chartes et manifestes éthiques sur l’usage des données ont ainsi été proposés au cours de ces dernières années, et nous proposons d’en faire un recensement partiel et subjectif ci-dessous.

Pourquoi OLKi ?

Saturday, Apr 20, 2019 by Christophe Cerisara AI, society

Problèmes actuels (constat) Les réseaux sociaux sont contrôlés par les USA et la Chine; c’est un problème pour la recherche européenne car: Ils contiennent l’information/la connaissance sur le monde la plus riche en langage naturel La communication scientifique passe de plus en plus par les réseaux sociaux “Data is the new oil” & “AI is the new electricity” Importance géostratégique des data + IA (mais sans data, pas d’IA) Les acteurs qui contrôlent les data contrôlent les progrès en recherche/science Méfiance et rejet grandissants des citoyens vis-à-vis de l’IA Il faut un autre modèle véritablement éthique Solutions apportées par OLKi Il faut intégrer aux recherches en IA la question de la gestion des data: nous proposons un modèle de gestion ouvert et équitable Les producteurs des data contrôlent leur data Ouverture = équité dans l’accès aux data API d’accès open-source et communautaire = contrôle partagé, possibilités d’expérimenter illimitées Réflexion interdisciplinaire sur l’éthique Insuffisance des “grands principes” de toutes les chartes éthiques actuelles, qui s’appuient sur le légal ou la confiance que les citoyens doivent avoir Il ne faut espérer une confiance des citoyens, mais leur donner les moyens de vérifier a posteriori: une licence open-source pour les modèles Rapprocher citoyens et scientifiques = amener la science vers les citoyens, sur les réseaux sociaux Vers un eco-système de recherche fédéré: Partage des coûts, extension sans surcoût, résilience… Emergence des communautés scientifiques et auto-structuration Intégration dans le réseau fédéré des data + communication + expérimentation Valeur ajoutée: Contrôle communautaire du réseau, des API (Expérimentations facilitées) Complémentaire des autres plateformes de stockage des data Intégration Data + Communication + Expériences (un seul réseau pour tous et toutes les modalités) Ethique, ouvert

A use case for the OLKi platform

Wednesday, Apr 10, 2019 by Christophe Cerisara AI, project

John is a linguist who is currently working on a specific syntactic structure, the coreference. He uses for this an open corpus of conversational French speech, taken from ORTOLANG. He decides to annotate part of this corpus with coreference structures, and he then wants to distribute his annotations also with an open licence. Beyond the dissemination of his work, John would also like to discuss with other linguists, as well as other scientists in general, who may be interested in these kinds of structures.

Deep Learning for scientists

Friday, Mar 1, 2019 by Christophe Cerisara AI, deep learning

The reference The deep learning book If you want to get into deep learning by studying, the academic way, then you should read this book. It does not include of course the latest recent developments (such as transformers), but it’s still pretty state-of-the-art. Video lectures & courses Best series of course into Machine Learning At least, these video courses by Nando De Freitas have my preferences because they are quite generic about Machine Learning (beyond deep learning) and are of very high quality.

Kickoff meeting

Tuesday, Jan 1, 2019 by Christophe Cerisara meetings, project

Our kickoff will be in Nancy on March, 15th, 2019

Blog

Contact