Unleash the full potential of your data
Artelys develops a data mapping tool with the French Ministry of Home Affairs to strengthen its data governance.
Improving passenger information for bus users in the Paris area: Artelys demonstrates the contribution of artificial intelligence methods to produce more accurate and stable estimated bus arrival times at their stops.
Artelys and the ANAP have been collaborating during 3 years to supply the healthcare sector with the most recent advancements in Data Science.
The MMT initiative (Man Machine Teaming) explores the possibilities of development for cognitive air combat systems.
Operating transport networks with the support of AI. This is the field explored by the online competition “Learn to Run a Power Network in a sustainable world“ launched by RTE with Artelys support.
A redesigned relationship between Man and Machine
MMT was launched and financed by the French General Direction of Armies (DGA). It is animated by the companies Dassault Aviation and Thales. Artelys is part of the ecosystem of partners, composed of French start-ups, SMEs and research centers.
As part of the Maestro project, Artelys is exploring the application of Monte-Carlo Tree Search algorithm and Reinforcement Learning to aerial trajectory optimization through Constraint Programming.
Official information: https://man-machine-teaming.com/
Data Science and AI to support crucial usecases in the healthcare sector
For 3 years, Artelys has been supporting the ANAP to implement techniques which solve concrete use cases. Those technics involve statistical data analysis, AI prototypes and have contributed to a strategy of internal acculturation to Data Science tools and Agile methodology.
Several use cases have been identified with the experts and have given rise to different services:
- Development and integration of a predictive module of the occupation rate of beds embedded in an existing tool for hospitals.
- Segmentation of the users of the web content of ANAP, according to their user profile. Statistics and clustering methods (K-Means) have been exploited to build efficient user personae.
- Analysis of the adequacy between healthcare professional demands and content offered by ANAP on its website. A list of clusters of interest have emerged thanks to a segmentation method (Hierarchical Ascendant Classification for biclustering) and led to recommendation concerning the structure of the offer.
- State of play of the real estate owned by the medico-social establishment in France. Creation and analysis of tailored indicators.
- Development of a configurable web tool using text mining to explore the thematic of Twitter posts relative to healthcare issues. Emergence of relationship between popular themes thanks to Natural Language Processing (NLP).
Gather the diverse available data streams to understand and act for road safety
The Loiret department is a member of the European project INTERREG NWE BE-GOOD and is in charge of the challenge “SAFER ROADS” whose goal is to bring characteristics by data use for road safety improvement. It means better understanding of the accident contexts and more efficient prevention actions.
One important stake is to get the data available and actionable by concerting the diverse actors of mobility sector who own it.
Artelys organized workshops in which several actors collaborated to provide data streams and specify their needs: public administration of the department (GIS, Road Observatory, Infrastructure administration), police forces (Gendarmerie Nationale), firemen (SDIS), insurance companies (Mondial Assistance, Thelem) and navigation services (Waze, Coyote).
A web platform has been developed by Artelys and made available to the public administration, stakeholders but also to citizens. It contains:
- A data warehouse with harmonized multi-source data streams and additional information on the context. This data are directly usable through requests or simple dashboard visualization.
- More than 24 000 accidents from the previous 10 years located on the road network and complemented by infrastructure and context information.
- 8 thematic dashboards to analyze influence of multiple variables on the accident risk (infrastructure, speed, traffic, seasonality, weather, accident situation, aggravating factors).
- A prediction algorithm which evaluates, on the road network, the risk level of accidents for a given future context. This algorithm is being developed and relies on supervised machine learning using decision trees.
- A clustering algorithm which has made some reference profiles of accidents emerged.
Operating transport networks with the support of artificial intelligence
In the competition ”Learn to Run a Power Network in a Sustainable World”, AI agents developed by the community have to face real-life TSO challenges.
They have to learn how to design the best set of actions on the grid in terms of robustness against realistic scenarios: consumption peaks, line overloads, intermittent renewable production, plants maintenance, but also cyber–attacks. Machine learning methods, reinforcement learning in particular, are designed to navigate in a vast multidimensional state space and converge to an optimal policy.
Artelys had several contributions at different levels of the competition, by developing four open-source Python packages.
- Grid2viz – An intuitive web application that offers interactive views into the results of Reinforcement Learning agents that ran on the competition environment (grid2op library). It has been developed with Dash.
- Chronix2Grid – A modular package that allows to generate synthetic but realistic consumption, production (solar, wind, hydraulic, nuclear and thermal), electricity loss (dissipation) and economic dispatched productions time-series for a given power grid. Parameters are tuned thanks to a set of indicators based on real-life time-series.
- ExpertOp4Grid – An “expert” agent dedicated to solve overloads in line grids using cheap but non-linear action with an expert approach inspired by recent research. For any new overloaded situations, it computes an influence graph around the overload of interest, and ranks the substations and topologies to explore, to find a solution.
- Oracle4Grid – An “oracle” agent that learns about the future by performing simulations and then finds the best course in order to maximize the cumulated reward. It provides indicators that provide benchmarks in agent performance.
See the competition on Codalab platform.
Data Science and Artificial Intelligence to improve the service provided to RATP passengers
Artelys has explored different approaches for the RATP to improve the short-term prediction of bus arrival at their stops. On the one hand, the optimization of the current model parameterization, and on the other hand, the exploration of innovative methods using state-of-the-art Machine Learning models.
Being able to offer the most reliable estimate of the next bus transit times is a major lever for the RATP in order to offer an optimal experience to its bus lines users. This forecast, which is by nature short term, is made difficult by many parameters. Some are structural, such as the number of intersections, the distance between stops, the day, time and/or week of travel, while others are contextual and need to be updated in real time, such as road traffic and weather conditions.
In addition to the current RATP system improvement by optimizing the existing algorithm, the use of predictive Machine Learning models has been studied. To this end, Artelys carried out:
- A review of the state-of-the-art methods applied to short-term bus arrival prediction
- The construction of data sets for model training: from RATP system logs and exogenous data such as rainfall and temperature
- An implementation of the different relevant AI models: linear regression, local regression (to include non-linear effects locally), Gradient Boosting, Random Forest, KNN etc.
At the end of this study, Artelys was able to identify that a mixture of a linear regression model with a Random Forest model brought a significant performance improvement, compared to the model currently used, both in terms of estimated bus arrival times accuracy and in terms of stability of the displays provided to users.
Artificial intelligence models are therefore a relevant lever to considerably improve the traveler information service provided to RATP users. The industrialization of the algorithms developed by Artelys must now be integrated into the development roadmap to make them available when the bus operators in the French capital are opened to competition.
A tool to centralize information
The French Ministry of Home Affairs wanted to strengthen the governance of its data by equipping itself with a solution for referencing the data sets available and their characteristics in order to enhance their value and, in particular, to accelerate data-related project initiatives.
Artelys is working with the French Ministry of Home Affairs to develop a web-based tool for mapping available data sets for use by all members of the Ministry. Based on Flask, React and Elasticsearch technologies, it enables the centralization of information knowledge that was previously dispersed among the various business departments.
Through this tool, users can explore (search, sort, filter) and then compare the various data sources referenced within the ministry. The mapping application offers a showcase of available data, facilitating the launch of new projects and the reuse of existing data already consolidated. In addition to these search functions, the tool integrates administration functions with access rights management by data perimeter. Each area has a manager, which makes it easier to monitor the relevance of information.
Evolutions on the current search engine are currently being developed. They will use Natural Language Processing methods. Indeed, these developments take into account the management of semantic fields and business nomenclatures. In addition, spelling and typing errors will be automatically corrected during searches.
The French Ministry of Home Affairs and Artelys have made the code open source, thus committing to the transparency of the tool and allowing its reuse by the other public administrations. It is available at the following address : https://github.com/dnum-mi/cartographie-donnees
From raw data to the right architecture
Artelys leverages its well-established software editing activity and the skills of its IT experts to create the strong foundations of your project.
Research & Development
20% of our activity is dedicated to R&D
Artelys is strongly involved in R&D, exploring and experimenting big data techniques and tools to provide its customers state-of-art solutions.
Time series modelling and projection
We offer our clients a better understanding of their activity’s trends, seasonalities and regime shift and a wide range of efficient models and tools.
Deploy the latest AI methods
Our data scientists design, develop, and deploy machine learning tools and algorithms that make your products and operations smarter.
Optimization power to enhance Datascience
The combination of DataScience and optimization offers extended perspectives in terms of accuracy and robustness for business problems resolution.
Your data strategy needs a partner understanding your business
Our experience in industrial sectors, Energy, Transport or Marketing combined with our empathy for your business issues allows for a deeper collaboration and the delivery of the most suitable solution.
Making visualization a business tool that your teams can use
Through maps, charts and simulations our ad-hoc graphical interfaces, helps to objectify your decisions on a quantitative basis.
Simulation & Risk Management
Be ready and prepared when the real thing happens
We deliver ’What if’ and Monte Carlo simulations to our clients, and implement the latest techniques of probabilistic forecast.
subscribe to our newsletters