Hosted projects

  1. Hosted projects
  2. Research IRIT teams studies

Hosted projects on OSIRIM :

SUMONN (IRIT (team SIG), Airbus, CERFACS) - february 2021

Machine learning of latent spaces on aeronautical data for reconstruction and simulation
The objective of the project concerns the creation and the implementation of surrogate models - "Surrogate Model" - based on neural networks, capable of learning the dynamics of a physical system on the basis of simulation data. Then, the model if designed will replace the "heavy" simulator, keeping the same level of fidelity.

Slant (IRIT (team Melodi), INRIA Lille-Nord-Europe (team Magnet), Université de Luxembourg (Lab: ICRST)) - december 2020

Analysis of linguistic bias in textual data.
The SLANT project aims at characterising bias in textual data, either intended, in public reporting, or unintended in writing aiming at neutrality.
An abstract model of biased interpretation using work on discourse structure, semantics and interpretation will be complemented and concretised by finding relevant lexical, syntactic, stylistic or rhetorical differences through an automated but explainable comparison of texts with different biases on the same subject, based on a dataset of news media coverage from a diverse set of sources. We will also explore how our results can help alter bias in texts or remove it from automated representations of texts..

Smart encoding of 3D-meshes using deep learning. Application to assisted design (IMT, IRIT (équipe Samova), ANITI, Safran Tech) - Décembre 2020

Design a tool for real-time estimation of Navier-Stokes equations around the blades of a turbo-compressor, using reduced order and surrogate models.
Turbine blade geometry is a very critical performance factor in aircraft engines, as they endure extreme temperature and pressure conditions. Therefore simulations involved in their design have to be very fine, which makes them very time-consuming. The aim of the project is to propose a deep learning-based approach to this issue, like a reduced order or surrogate model able to produce an estimation of the RANS solution around the blades, in order to optimize their geometric shape.

Multilingual Automatic Survey generation (IRIT (team IRIS)) - december 2020

This project aims to address the problem of automatic multi-lingual abstract/survey generation and deals with the use of scientific literature as input.

Langues Peu Dotées (IRIT, MISA Univ Antananarivo (Madagascar ), Addis Ababa University, (Ethiopia)) - october 2020

Tools and resources for processing texts in poorly endowed languages.
this involves creating automatic text processing tools (indexing, rooting, etc.) for under-equipped languages such as the languages of Madagascar (Malagasy) or Ethiopia (Amahric).

CIPAD (IRIT , Institut National Universitaire Jean François Champollion, ISIS Castres) - october 2020

The objective of the project is the implementation of a connected cane for the elderly. This rod carrying sensors and communicating via a wireless link that consumes little energy has the following objectives: (1) The transmission of high resolution data on the evolution of the user's gait (2) The facilitation of the interfacing between the user and the smart city ecosystem.
The information collected from the cane is primarily intended for a personalized and scalable fragility prediction model: it aims to detect the imbalance of gait via a combination of signals from on-board sensors. A prediction model will be developed using machine learning and deep learning methods. This model will be able to predict the evolution towards the fragility of a cane user. This information may be made available to nursing staff so as to enable them to guide the elderly person in a relevant way in their care path. This project also covers network aspects which are not detailed in this paragraph.

LEARNFRAM (IRIT - INPT - TTT) - august 2020

The project aims to create a general framework for processing medical signals and images to detect bio-medical anomalies and risks. In addition to the framework, the project creates an application for monitoring hypovigilance. The objective is to promote research results by creating a software demonstrator.

neOCampus internship (IRIT - UPS) - july 2020

The project objective is to simulate a large scale RPis network. Each running an AI algorithm to make a decision: cache the data it sees passing through the network or not.
The purpose of the internship is to develop new approaches to routing and collecting data in the IoT infrastructure of neOCampus.

PREVISION (IRIT team SIG) - june 2020

The project is linked to the PREVISION project funded by Europe under contract N ° 833115. This project brings together technical partners and users. The platform developed aims to help organizations in charge of security and to detect radicalities and their level.
In this context OSIRIM will be used to develop the programs and verify the results on different test collections, outside the project and within the project. In particular a jargon extraction module, a module to complete missing data and a radicalization detection module will be developed by IRIT-CNRS

SIG-COVID-19 (IRIT team SIG, LSIS) - june 2020

COVID-19 data mining and information retrieval
The core corpora are (a) a corpus uploaded by us on PubMed (b) an open COVID-19 corpus
Several tasks will be carried out in parallel on one or the other of these corpora:
- Participation in TREC Covid-search
- Data mining for the extraction of salient terms
- Multi-document summaries on specific subjects (example: the origin of the covid)

Transparens (IRIT team SIG, CHU Toulouse) - may 2020

Enhancement of medical data on brain injuries through several technologies (classification of brain injuries from MRI images)

Clone (IRIT, Serious Game Research Lab / Institut National Universitaire Champollion) - may 2020

Search for anomalies in the course of nursing students in the clone serious game.
Several methods will be tested and carried out for the analysis of data from clone game paths carried out by student nurses.

C2C : Car to Cars (IRIT, CLLE, LIRMM, OKTAL company) - may 2020

The C2C project aims to study the innovative issue of collective autonomous and connected vehicles as well as the acceptability and appropriation of the human user of these vehicles and of the human moving in the environment populated by this type of vehicles.
This project is therefore studied according to two complementary axes: 1) ergonomic and societal and 2) IT. From an ergonomic point of view, it aims to provide an architecture and recommendations favoring the acceptance and appropriation by humans of these vehicles and based on the study of indices determining the behavior of humans (driver and / or passenger). From an IT point of view, it is a question of studying how each vehicle in the collective communicates with its neighbors and behaves / reacts in response to the information provided by its neighborhood. A lock notably consists in determining the information which is relevant to communicate among all the data retrieved from the vehicle sensors. The project has an IRIT physical simulator running the SCANeR Studio software platform, from which driving simulation data will be extracted and processed using, for example, machine learning methods

DeepRad (ANITI, Cerco, ISAE-SUPAERO, IRIT Team SC, NXP company) - april 2020

Multiple target extraction, identification and tracking for automotive radar
The aim of the project is to build new machine / deep learning frameworks specifically tailored to radar data, that will be incorporated in real automotive systems. Those algorithms are expected to detect, identify and track surrounding objects in complex environments like cities or large cross-roads, traffic jams, etc.

SGL Thresh (CERCO - IRIT SAMOVA teams) - March 2020

SGLThresh: Surrogate Gradient Learning of Thresholds.
It is a question of fixing optimal thresholds for a multi-label classification problem. Thresholds allow you to assign a label or not based on continuous probabilities. We want to find the thresholds which minimize a certain cost function by using a gradient descent. The thresholding function being non-differentiable, when calculating the gradient it is claimed that a sigmoid was used instead.

REAM (IRIT ADRIA and MELODI teams) - January 2020

Explicit Reasoning and Learning on Medical Imagery.
The project objective concerns the validation of neural network architecture for the automatic classification of histopathological images, and verification of learning incrementality hypotheses.
This project is part of a thesis on the interaction between the learning system and the explicit reasoning system. The first part of the thesis aims to design an automatic image classifier using deep learning. In a second step, explicit knowledge will be added in the form of ontologies or expert system rules. Thirdly we will develop the interaction between the two systems.

ANITI - DAML (Université Fédérale de Toulouse MIP) - November 2019

Data assimilation (DA) is a proven technique for monitoring a dynamic system through the use of a parameterized representation of its dynamics and physical observations. It is traditionally presented in a Bayesian context.
Supervised Deep Machine Learning (ML) allows for weight-wise label prediction of a neural network using training examples.
The central question for the Chair is: how can we introduce an approximate dynamic into neural networks? Conversely, how can we introduce deep learning in AD to be able to do without dynamics and obtain better algorithms. In both cases, what are the expected performances?
The purpose of the Chair is to propose hybridization algorithms by validating them on physical equations representative of real equations..

Hearvolution (IRIT team REVA - AMIS Laboratory UT3) - October 2019

Identification of positive selection signatures in genes involved in auditory processes at the interspecific scale in human and non-human primates.

Toulous School of Economics - October 2019

GPU parallelization of codes for the study of a discrete choice demand model for food economics.

Nurisens (IRIT Team SIG - INSERM I2MC - Innopsys, Keleo Solutions, Flutilliant companies) - September 2019

The concept is based on a urinary self-test connected to a mobile phone and a data platform gathering anonymised data and achieving data correlations.
IRIT will focus on developing algorithms for the analysis and correlation of both individual and collective data by user groups in order to identify and/or anticipate drifts of nutritional behaviours in the objective of suggest recommendations for nutrition behaviours.

DIAMEN (IRIT Team SIG - CHU Toulouse - ORME Company) - July 2019

Development of a diagnostic aid tool in cardiac scans of nuclear medicine departments.
This involves developing algorithms from Machine Learning and Deep Learning to build an AI capable of predicting sick patients according to different cardiac pathologies. The first architectures developed will be based on convolution neural networks (CNN)

aireps (IRIT Team MELODI - CERCO Laboratory) - May 2019

Analysis of semantic representations, comparison of human and AI representations.

dfo4rl (IRIT Team APO - Lehigh Université (USA)) - May 2019

Development of no derivative optimization algorithms for reinforcement learning.

Neocampus (IRIT) - April 2019

Training of neural networks for smart sensors. First applications: attendance measurements, detection of Asian hornets at the apiary, ...

Flux Doppler (IRIT Team Minds - Université de Tours CHU - April 2019

Measurement of blood flow from Doppler signals by ultrafast acquisition

COOSINET (IRIT SC Team- CNES (Toulouse), Thales (Cannes)) - March 2019

Compression Onboard Of Satellite Images using neural NETworks

Text mining project on scientific literature (IRIT IRIS team - INIST) - March 2019

The project aims to apply text mining techniques (in particular word embeddings) on large corpuses of scientific literature (in the order of 20 million texts on ISTEX, for example)

LINTO (LINAGORA company - IRIT (SAMoVA and MELODI teams) - LAAS - Zelros company - LIX (Ecole Polytechnique - Saclay)) - March 2019

Conversational assistant working in a professional environment

COST ANR Project (IRIT - CLLE (Toulouse), LIG (Grenoble), LIP6 (Paris)) - March 2019

Modelling Complex Tasks in Information Research

Image CLEF 2019 (IRIT - Univ. Toulouse 3) - February 2019

Deep learning for medical data and participation in the CLEF challenge

Callisto CALMIP - February 2019

Tools for Dataset Description

Discourse analysis via clustering methods IRIT - October 2018

The project aims at extracting knowledge from a discourse database and at comparing different methods from supervised classification (neural networks) and clustering (spectral clustering)

Logical Frameworks for Abstract Argumentation (LF2A) (IRIT - CIMI) - May 2018

Deforestation (IRIT - Univ. Madagascar) - February 2018

Applying machine learning algorithms (CNN) on satellite images for annotation and change detection.

Victoria (UPS - INP / IRIT) - January 2018

Projet H2020 – SEC (Video analysis for Investigation of Criminal and TerrORIst Activities) for the dissemination of large reference corpus (several Teras) between the different teams in Europe.

Depression (IRIT - LIMSI) - March 2017

The aim of the project is to detect depressed people via their posts on social networks by first using CLEF's e-risk collection and then other collections.

LisTIC (LISST - IRIT - LERASS) - November 2016

Sociological investigation on the uses and media effects of social networking websites (Facebook and Twitter) on personal social networks and online participation during French presidential elections and Brazilian social movements.

Parolothèque (CHU Toulouse - IRIT - IFERISS - OCTOGONE - CLEE - LIA - LPL) - November 2016

The main goal of this project is to establish a body of voice for cancer patients. This corpus can then be analyzed and studied as much from the point of view of its container (speech) as its content (speech). It will develop various research areas and the development of scientific publications in various fields (humanities and social sciences: SHS, epidemiology, linguistics, computer science ...).

MUSK (LISIS) - June 2016

The MUSK Project is dedicated to textual data analysis (Text Data Sciences) with different type of data (texts, microblog, audio and videos) issues from several sources (technical documentation, scientific publications, social media.

Learning to Rank and Best System Configuration (Université Toulouse Jean Jaurès - Université de Montréal - IRIT) - February 2016

From several thousand system configurations that process queries, the goal is to learn the best configuration of a system for a given query. Once learned from a set of examples, the system must be able to handle any new query optimally.

Tweet Contextualization (IRIT - Université d'Avignon) - January 2016

Tweet contextualisation around events.


Meta mining for biomedecine recommandation.

PETASKY (LIRIS - LIMOS) - December 2015

This project deals with the problem of management of scientific data in the field of cosmology, more particularly for the data collected within the framework of the LSST project ( Experiments on data partitioning techniques were conducted on the OSIRIM platform, in the form of Jobs MapReduce.

POLEMIC (IRIT SIG - Departamento de Tecnologías de la Información, Universidad Autónoma Metropolitana (UAM-Cuajimalpa), Mexico City) - September 2015

Analysis of users' behavior in social networks, in particular the temporal evolution of this behavior through the interactions of the user in his network.

IRIM-at-Trecvid (Consortium IRIM (LIG, LISTIC, LaBRI, CEA LIST)

Experimentation of several French scientific teams as part of their participation in the international TRECVID campaign.


Interactive and intuitive editing and control of synthetic images.

SemDis (CLLE - ERSS  et IRIT)

Creation and evaluation of French distribution bases.


Aggregative data search.


Research IRIT Teams studies on OSIRIM :

SIG Integration, NoSQL, retrieval information, analyzing, mining and visualizing information.

IRIS Indexing and searching for information in large volumes of texts to notably contribute to international benchmarks in information search : TREC, CLEF, INEX, YANDEX ...

SAMOVA Evaluation of tools for indexing musical content, indexing large volumes of international television program recordings.

IMTstat Experimentation of distributed statistical computing techniques for large masses of data.

MELODI Textual corpora analysis and ontologies.

TCI Complex image processing.

Reva Use of mathematical and algorithmic methods for the analysis of multimodal, visual and biological data.

Pyramide Dynamic optimization of large-scale distributed queries.

SMAC Cooperative Multi-Agent Systems.

SIERA Detection of network intrusion & weak signals

SC Signal & Communications


Previously Hosted Projects:

: innovation and industrial research on automatic analysis and enrichment of digital, multimedia and multilingual content, hosting tasks placed under the responsibility of IRIT teams.

    Owner : SIG  and  SAMOVA