SLIDE Current Funded Projects
Adaptative Learning for Intelligent Crowdsourcing and Information Access
Funding: Contenus numériques et interactions, ANR 2013
Participants : S. Amer-Yahia, V. Leroy
Partners: AlephD, Telecom ParisTech, LRI (Univ. Paris Sud), IMT(Univ. Paul Sabatier), Vodkaster, Xerox
The target of this project is the development of methods for information access and intelligent crowdsourcing in collaboration with Université Paris Sud, LTCI, Xerox, and UPS/IMT. In the context of information access (e.g. search or recommendation), building and maintaining user preference profiles helps applications satisfy diverse preferences. For intelligent crowdsourcing (e.g. data sourcing and micro-task completion), expertise profiles help better assign task to users. In both scenarios, the key challenges are that user preferences and expertise cannot be known in advance; and can rarely be explicitly declared by uses in a reliable or stable way. Consequently, preferences and expertise need to be discovered over time via a learning approach. Our project’s goal is the study of models and algorithms that rely on adaptive learning techniques to improve the effectiveness, performance, and scalability of user-centric applications.
Funding: Programme AGIR (Grenoble INP, University Joseph Fourrier)
Participants : S. Amer-Yahia, V. Leroy
Enable health-related hypothesis testing between physical and virtual spaces at a large scale.
Funding: CNRS Mastodons 2014
Participants : S. Amer-Yahia, N. Ibrahim
Partners: Université Paris 13, Université Paris Descartes, CNRS EVS (Environnement, Ville et Société)
The goal of the CrowdHealth project is to enable health-related hypothesis testing between physical and virtual spaces at a large scale. CrowdHealth aims to establish a collaboration between researchers on scalable algorithms for mining personal data and times series, nutritionists, and researchers studying individuals in geographical spaces. CrowdHealth will rely on Nutrinet and Twitter as data sources for extracting and verifying correlations between demographics, nutrition and health. It will also leverage Crowd4U, an academic crowdsourcing platform to label data and validate extracted hypotheses.
Découvrir, expérimenter, mettre en œuvre et appliquer le Big Data
Funding: Investissement d’Avenir
Participants : S. Amer-Yahia
Partners: Business&Decision, Groupement des Mousquetaires, INRIA Saclay, LIFL, LIRMM, Eolas
The aim of this project is to develop scalable algorithms for data mining and processing in collaboration with INRIA Saclay, LIFL, LIRMM and industrial partners: Eolas and B\&D. The project defines 3 use cases, all of them with industrial impact. The first use case, network analysis applied to data centers datasets, aims to provide interactive traffic monitoring interfaces including traffic aggregation over time abnormal traffic identification. The second use case, digital marketing, applied to server and application logs, aims to provide customer-centric statistics and customer engagement analysis using sequence mining. The third use case, linked open data, aims to develop a platform that integrates open data on the city of Grenoble and makes it readily available for the development of various applications.
The Living Book of Anatomy to Visualize the working/acting body using Augmented Reality techniques.
Funding: Persyval Lab
Participants : F. Jouanot
Partners: TIMC, GIPSA-lab, LJK, LIG
Through the Persyval team we propose to develop the « Living Book of Anatomy » also
called LBA to visualize the working/acting body using Augmented Reality techniques. The
idea is to capture an user action (limb movement, speech sound or orofacial movement) and
to display the muscles responsible for this action and their activation, by superimposing
this information onto the user’s body (Augmented Reality). This tool could be used to
facilitate learning anatomy for medicine students and to provide an interactive way for
general public to discover human anatomy in action. The assumption underlying this work
is that experiencing the anatomical knowledge through body action, or embodiment,
could make the acquisition of this knowledge easier. We aim at evaluating this hypothesis
as well as the interactive designed techniques with medicine students. A large collection of
accessible data and applications describing anatomy or other medical knowledge are now
available, based on advances in numerical sciences and more specifically to the spread of
smart phones and tablets in general and more specifically in the clinical world. However, as
far as we know, none of these new services and applications bridges the gap between the
user’s own body and medical knowledge. A core hypothesis of the LBA project is that the
use of Augmented Reality could make the embodiment of knowledge easier by a closer
connection between models, knowledge and reality.
Developing the LBA requires the resolution of several scientific or technical issues : (1)
capturing the user’s actions, (2) mapping the model to the user’s body and conversely,
(3) rendering anatomical and functional information onto the user’s body (4) designing
augmented reality interaction techniques for interaction with the physical (user’s body)
and digital information (anatomical/functional displayed information) on a mobile device.
These issues can be addressed with different levels of complexity. First, a suitable
knowledge base must be built, using Ontoly-Based Database Access, including medical
knowledge of the description of anatomical structures and organs, together with their
functions and behavior (including biomechanical models, medical images, etc.). Second,
Interactive techniques should be developed to make the information handable by the
user, with several levels of refinement and expertise. One challenge is also to make the
LBA system able to render such information in real or interactive time. Third, the approach
should be evaluated, which will require to understand the cognitive aspects and more
specifically the link between knowledge and body-experience (see references here after).
Mining data streams with hardware support
Funding: Persyval Lab
Participants : V. Leroy
Partners: STMicroelectronics, TIMA, LIG
Data production grows continuously. The development of the Internet of things and sensors produce terabits of activity traces. Pattern mining algorithms are a cornerstone of data mining. They consist in detecting frequently occurring patterns (sets, sequences, sub-graphs) In data. These patterns can later be used in many applications, including classification, recommendation and prediction. Existing approaches to pattern mining focus on batch processing, that is offline computation. However, more recent work considers stream (online) processing. Stream processing has the advantage of reading data only once, which limits the complexity at the cost of approximate results. They also allow continuous analysis, hence results can be obtained with low latency to detect anomalies in real time.
The goal of the Nano2017 ESPRIT project is to propose a hardware solution for pattern mining in high throughput data streams. This solution, which could be proposed as a support hardware card, will be able to test simultaneously the presence of a large number of patterns in the data. The benefits of such a solution are (i) to process faster streams than purely software approaches, and (ii) to use less servers to process data streams, thus reducing energy consumption.
Practical Algorithms for Ontology-based Data Access
Funding: ANR 12 JS02 007 01, Programme JCJC
Participants : M.C. Rousset
Partners: LRI, LIRMM, LADAF
The aim of this project is to develop practical algorithms for ontology-based data access (OBDA) in collaboration with LRI, LIRMM, and the LADAF (Laboratoire d’anatomie de Grenoble). This project is centered on two challenges:
(i) Scalability: in contrast with relational database management systems that benefit from decades of research on querying algorithms and optimizations, ontology-based data access is a young area of study, and despite important recent advances, including the identification of interesting tractable ontology languages, much work remains to be done in designing scalable OBDA query answering algorithms.
(ii) Handling data inconsistencies: In real-world applications involving large amounts of data or multiple data sources, it is very likely that the data will be inconsistent with the ontology, rendering standard querying algorithms useless (as everything is entailed from a contradiction). Appropriate mechanisms for dealing with inconsistent data are thus crucial to the successful use of OBDA in practice, yet have been little explored thus far.
Quality and Interoperability of large Documentary Catalogues
Funding: ANR – CONTINT 2011
Participants : M.C. Rousset
Partners: LIG,LIRMM, LRI, ABES
The aim of this project is to develop methods and algorithms for improving the quality and
interoperability of large documentary catalogues in collaboration with LIRMM, LRI, ABES
and INA. The approach chosen in this project is a knowledge representation approach
based on representing data using Semantic Web standards. This allows on one hand to
give a logical semantics to the notion of quality and, on the other, to use reasoning
mechanisms for data linkage. The objectives of the project are (1) the development of a
quality model for the individual entities (named entities) identification problem, (2) the
definition of a trust model suitable for data reconciliation and fusion and (3) the discovery of
entity identification characteristics and their manipulation by different techniques. A large
part of the project is devoted to the evaluation of the proposed approach by experiments
conducted on suitable test benchmarks and the development of demonstrators adapted to
the two document databases owners involved in the project.
Methods and tools to process execution traces.
Funding: FUI-Minalogic, OSEO
Participants : A. Termier, F. Jouanot, N. Ibrahim
Partners: INRIA, LIG, TIMA, STMicroelectronics, Magilem, probayes
The SoC-Trace project aims to develop a set of methods and tools based on traces of execution produced by multi-core embedded applications. It will allow developers to optimize and debug these applications more efficiently. Such methods and tools should become a building block for the design of embedded software, in response to the growing needs of analysis and debugging required by the industry. The technological barriers consist of a scaling problem (millions of events stored on gigabytes) and a trace understanding problem related to applications whose complexity is increasing. The project addresses the problem of controlling the volume of tracks and of developing new analysis techniques. SocTrace is composed of academic partners with related themes, and several industry partners including STMicroelectronics.