MAMUT projects (Machine learning And Matheuristics algorithms for Urban Transportation) aims to set up an open and open-source platform for solving recurring, dynamic vehicle routing problems in urban environments. These are large-scale logistics problems. The scientific and technical objectives concern the resolution of single and multi-vehicle routing problems, but also the dynamic aspect involving the real-time consideration of new demands. Three challenges are presented, including two scientific challenges (characterization of urban logistics instances, generic RO/IA hybrid solver) and one technical challenge (platform implementation). The main original feature is the hybridization of operational research (OR) with artificial intelligence (AI)
PEPR eNSEMBLE 2023eNSEMBLE is an exploratory PEPR (Priority Research Program and Equipment) funded as part of the France 2030 future investment plan. The eNSEMBLE project (Future of Digital Collaboration) aims to fundamentally redefine digital tools for collaboration. The targeted PC4-Congrats project aims to better manage very large knowledge-producing collectives such as Wikipedia. The aim is to understand how they work (for example, why people come, contribute, share, stay or leave), improve collaboration (performance, quality, well-being, results, etc.) by providing players with useful and usable tools. Through the DECIDE team's Complex Networks axis and with a multidisciplinary partnership, we will mobilize different Complex Networks-based approaches to study the mechanisms that govern the growth, structuring and sustainability of these large online collectives. In particular, we will study inter-team dynamics, involving algorithms for community detection, anomaly detection, pattern detection, and new organizational metrics for understanding large real graphs.
Title: The effects of Artificial Intelligence on police activity: new quantification regimes, market diversification and redefinition of urban security systems.
Keywords: Public safety policy, urban government, artificial intelligence, explainable artificial intelligence (XAI), time series, bias.
Intelligent video surveillance, facial recognition, predictive mapping, etc. are just some of the tools that are renewing police activities. They are supposed to build "safe cities" using artificial intelligence (AI). This research project is studying different cities using these technologies in order to measure and improve the practical effects of AI on police activities. In this interdisciplinary project (sociology and computer science), the objectives are to: (i) Assess the organisational changes in police activities as a result of these innovations. (ii) Characterise the relationships between the different stakeholders (scientific, industrial, public and police) in the 'safe city' and identify the technologies that are likely to meet police needs and uses, and the mutual influences that these collaborations may have on the type of AI research and police activities. (iii) Correcting and completing digital data sources (from sensors, statistics, video surveillance, etc.). By contextualising them and highlighting certain biases, the contribution is to improve the reflexivity that police can have about its own practice, by proposing a new AI approach based on explainable machine learning.
Partners: IMT Atlantique, Université de Versailles-Saint-Quentin-en-Yvelines, Sciences Po Saint-Germain-en-Laye
Past projects
Projects we have coordinated:
Atypical project, funded by TSN Carnot (2023)
This project was carried out by Esteban Bautista, postdoctoral fellow co-supervised by Cécile Bothorel, Laurent Brisson (DECIDE team) and Grégory Smits (MOTEL team). The objective of the project is threefold: 1) identify and understand so-called regular behaviors in a potentially large interaction graph, 2) detect anomalies by taking into account structural properties (unconventional or unacceptable interactions), temporal properties (frequency of interactions) and spatio-temporal (change of affiliations to a community or sudden and massive modification of the structure of a community for example), and 3) explain the criteria which make an interaction scenario judged suspicious and thus helping decision-making actors to differentiate anomalies and emerging phenomena.
Related publication: MAD algorithm (publication in conference WSDM 2024), a statistical model modeling the interactions of a link stream through a random process.
The LEARN-IA project seeks to improve the energy performance of infrastructure in industrial and local authority environments using artificial intelligence and innovative data acquisition tools. The project involves improving the predictive analysis and recommendations provided to energy administrators using contextual information that is collected and then interpreted using semantic algorithms. The project also aims to improve the information retrieval process using technologies that enable the acquisition of non-measurable events (maintenance events, visual faults and anomalies, problems described by text, etc.) using systemic approaches:
capture these events in natural language using a suitable ergonomic input tool, by associating them if necessary with one or more physical digital measurement elements,
analysing and structuring this information using natural language processing (NLP) algorithms
extract knowledge from the links established between events and measurements using machine learning algorithms, and provide an explanation of this knowledge.
Keywords: time series, text analysis, software, XAI (explainable AI) Partners: IMT Atlantique, Energiency, Script&Go Major Publication: Ikram Chraibi Kaadoud, Lina Fahed, Tian Tian, Yannis Haralambous, Philippe Lenca. Automata-based explainable representation for a complex system of multivariate times series. IC3K 2022: 14th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management - KDIR, Oct 2022, Valletta, Malta. pp.170-179, ⟨10.5220/0011363400003335⟩.
MySmartLife (H2020) (2016-2022)
The mySMARTLife project aims to make the three flagship cities of Nantes, Hamburg and Helsinki more environmentally friendly by reducing the cities' CO2 emissions and increasing the use of renewable energy sources.
As part of this project, the DECIDE team is in charge of proposing a methodology and a multi-criteria decision-support tool for planning the upgrade of the Ile de Nantes heating network, involving the network's various stakeholders and decision-makers.
MEDISA is a research and development project aimed at creating a multi-criteria decision-making tool for sizing wastewater treatment systems. This tool is designed to take into account regulatory requirements, costs and environmental impacts.
The project is being led by a multi-disciplinary team comprising professionals from the wastewater treatment, modelling, measurement and ecotoxicology sectors, as well as academics (Eau du Ponant, IMT Atlantique (Lab-STICC), UBO (LMBA), LABOCEA, ACRI-HE, 3DEAU). The results of the project will enable those involved in wastewater treatment to make more informed decisions and contribute to environmental protection.
The DECIDE team has proposed a methodology for evaluating sanitation system improvement scenarios and implementing it in a software.
This research project on online information pluralism aims to analyze and assess the socio-economic effects of transformations brought about by digital technology on information quality and pluralism (IQP) in the media universe. PIL is a collaborative, interdisciplinary research project bringing together researchers in economics, sociology, computer science, linguistics, law and information and communication sciences.
As part of this project, the DECIDE team is working in particular on :
Analyzing interaction dynamics by developing algorithms for monitoring the evolution of communities in complex networks, so as to describe pluralism of opinion and, in particular, detect and describe the dynamics of echo chambers.
Algorithmic definition of the "quality" of a series of documents and of the "pluralism" of a given medium, using automatic natural language processing approaches.
A complete chain of automatic (or semi-automatic in some cases) analysis of document series, leading to intra-medium and/or inter-medium analyzers and rhetorics.
Academic partners: IMT Atlantique, Sorbonne Nouvelle - Paris 3 and La Rochelle Universities.
From sensors to knowledge: Communicate and decide
