Heterogeneity of data and methods: A unified collaborative framework for interactive temporal data analysis – HERELLES
HERELLES: Data heterogeneity - Method heterogeneity: A unified collaborative framework for the interactive analysis of temporal data
This project aims to design an approach that allows collaboration between different learning algorithms through the transfer of information from the analysis of complementary temporal data. This collaboration will be controlled by the expert who will intervene by evaluating intermediate results, and by injecting or validating new information (examples, constraints, knowledge...) proposed by the system.
Challenges and assumptions
Breaking with the current approaches each based on a single analysis paradigm, the HERELLES scientific project proposes to :<br />- Define a generic architecture allowing multi-paradigm methods (supervised vs. unsupervised) potentially working on different data to collaborate and to define the optimal conditions for its use (H1); <br />- Develop mechanisms for interaction with the user, offering him the possibility of injecting new information and reducing the semantic gap between the results and the expert's intuitions (H2);<br />- Propose methods for extracting and capitalising on the knowledge directly or indirectly produced when using the active collaborative method (H3);<br />- Implement and validate the proposal in the context of heterogeneous time series analysis (H4). <br /><br />The proposed method will be generic and will not depend on the potential fields of application. In order to validate its operability, we will be interested in understanding complex phenomena in our environment (land artificialisation, urbanisation, infrastructure construction, etc.), mainly via heterogeneous satellite data, a field in which the expertise of TETIS and ICube is proven. <br /><br />In addition to the methodological contributions, HERELLES aims to produce an operational platform for the validation and perpetuation of the proposals but also as a showcase for the project. It will complement the existing FODOMUST platform that ICube has been maintaining for over 12 years by integrating our proposals
In addition to a task T0 dedicated to the coordination of the project, we have structured the scientific programme in 4 tasks.
Task T1: Multiparadigm collaboration
Coordination: A. Cornuéjols & C.Wemmert - Main partners: AgroParis (31 p/m); ICube (25 p/m),
Objectives: To propose a new operational architecture and a theoretical framework to make supervised and unsupervised learning techniques collaborate in an iterative procedure (T1.1). To study the conditions for the implementation of such an architecture. Define methods to optimise the process and ensure process convergence and prevention of negative collaborations (T1.2).
Task T2: Incrementality and Interactions
Coordination: T.-B.-H. Dao & S. Loudni - Main partners: LIFO (28.4 p/m); Greyc (20.8 p/m)
Objectives: To design efficient mechanisms for incrementally integrating expert constraints while limiting consequent perturbations and taking into account potential inconsistencies between them (T2.1). Design an active interaction method by proposing mechanisms for selecting the information/questions to be submitted to the expert and optimising the expert's feedback (T2.2).
Task T3: Knowledge capitalisation
Coordination: B. Crémilleux & M. Roche - Main partners: Greyc (26 p/m), TETIS (14), LIFO (12)
Objectives: To design original mechanisms for the semantisation of clusters, the explanation of new clusters or new evolutionary profiles of a set of clusters (T3.1). Propose new ways of identifying and learning the user's needs and preferences in order to capitalise on the knowledge resulting from the data and interactions between the system and the expert in order to adapt its environment to his know-how and context of use. Implementation of mechanisms for generating constraints that can be exploited by the methods defined in task T2 (T3.2).
Task T4: FODOMUST platform and validation
Coordination: D. Ienco & P. Gançarski - Main partners: TETIS (20 p/m), ICube (23 p/m)
Objectives: Methodological validation of the various partners' proposals (T4.2) and thematic validation in the specific field of Earth observation (T4.3). Sustainability of the methodological proposals through an operational platform. Capitalisation and extension of ICube's work on a user-friendly interface allowing interaction with the process: visualisation and explanation of results, input of new knowledge (T4.1).
-- Ongoing --
-- Ongoing --
-- Ongoing --
In a break with current approaches, each based on an analysis paradigm, the HERELLES scientific program aims to define an operational theoretical framework that allows the combination and collaboration of different analysis paradigms and allows strong interaction with the expert for the extraction of knowledge from heterogeneous multi-temporal data. We aim for an approach that will allow collaboration between supervised or unsupervised methods through information transfers (clusters will be used to create learning data or to increase their size, learning data will be used to help validate and thematize clusters...) extracted from different but complementary time data. This collaboration will be controlled by the expert, who will intervene to incrementally add new knowledge. Thematic validation in remote sensing.
Project coordination
Pierre Gancarski (Laboratoire des sciences de l'Ingénieur, de l'Informatique et de l'Imagerie (UMR 7357))
The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.
Partnership
TETIS Territoires, Environnement, Télédétection et Information Spatiale
ICUBE Laboratoire des sciences de l'Ingénieur, de l'Informatique et de l'Imagerie (UMR 7357)
LIFO EA 4022 LABORATOIRE D'INFORMATIQUE FONDAMENTALE D'ORLÉANS
GREYC Groupe de recherche en Informatique, Image, Automatique et Instrumentation de Caen
MIA Mathématique et Informatique Appliquées
Help of the ANR 739,546 euros
Beginning and duration of the scientific project:
November 2020
- 48 Months
