We enable the integration of artificial intelligence with complete confidence, especially when the stakes are high.
Given its unique know-how, Numalis is involved in a variety of national and European high-stakes projects. The aim is to take on challenges, each one pushing further the boundaries for civil and military AI applications.
Supported by INRIA, NoLeFa-84 is a two-year pilot project that starts at the end of 2024 and aims to lay the foundations for Union AI Testing Facilities that will support the application of the EU AI ACT through testing, coordination and standardisation efforts.
GENIUS is an EDF project supported by GMV. This is a more comprehensive and advanced ‘follow-up’ to the CONVOY project on the detection of improvised explosives using data fusion (ground, air and space) and artificial intelligence.
A major European Defense project led by GMV, the CONVOY project will combine sensing technologies for the detection and recognition of hidden threats, including improvised explosive devices (IEDs) and landmines. The combinations of intelligent robots, drones and sensing technologies will utilise tactical cloud infrastructures and explore artificial intelligence to detect, recognise and avoid/neutralise hidden threats. Numalis will help with the design of the AI and help to assess and improve its reliability.
Funded by a Readynov grant of Region Occitanie, the RAHW project (Robust AI on Hardware) aims at implanting neural networks trained on desktop computers onto specialized embedded hardware such as FPGA. The challenge is to maintain the robustness of the networks while changing the arithmetic of the machine (fixed-point instead of floating-point arithmetic) and while limiting the word-length of the numbers. In short, implanting a neural network on an FPGA is a challenge comparable to fitting an elephant into a bottle. RRRAHWWW !
This European project aims to create a trusted AI evaluation environment for ML with a view to developing an aeronautical certification framework for these algorithms. The applications of the environment developed will be exclusively linked to machine learning evaluations relating to safety in all the areas covered by the EASA’s basic regulations.
In the aeronautics industry, the challenge is the reliability of Machine Learning and Deep Learnig solutions. Guarantees are sought on the “generalization of the machine learning model”, the “exhaustiveness and representativeness of the data”, and the “robustness of the algorithms and models”.
Major European defense project led by Leonardo on drone detection and implementation of countermeasures.
JEY-CUAS covers study and design activities: capability gaps analysis, CONOPS definition and Preliminary Requirements Review (PRR), System Requirements Review (SRR) and Preliminary Design Review (PDR). The PDR is supported by partial tests illustrating and assessing the criticality of innovative and promising concepts and technologies as well as of the overall system design for risk reduction.
Major European defense project led by Airbus to demonstrate the benefits of AI for better situational awareness, decision-making and planning.
The project covers both a functional approach (to derivate military needs in functional capabilities) and a technology assessment to match these functional capabilities. On the basis of already existing solutions and mature technologies, a gap analysis is performed to assess trends for performance enhancements to address future scenarios.
The aim of this project is to design a chain for improving the generation of synthetic data. Oktal SE provides the generation of synthetic data and Saimple is used to assess the robustness of the model trained with as well as providing information to aid explainability.
The aim of the project is to work on detectors and help the SNCF design a methodological document for validating RNs.
During this project, AIs for the detection of railway signaling panels that could be embedded in autonomous trains are being analyzed. The aim is also to provide proof of reliability and a validation methodology with a view to acceptance of the use of AI in railway standards.
Sponsored by MBDA, this BPI challenge aims to help validate AI algorithms based on the SVM method and/or neural networks for Defense purposes and to improve the understanding of their overall robustness over the input domain.
Along with University of London, ENAC and Spinworks, the aim of the project was to identify mathematical approaches to support the design and the verification of a next-generation AI-based Guidance, Navigation and Control (GNC) module for space exploration. The study makes a formal link between the AI-based machine learning (ML) and the control theory-based reasoning and optimisations within a challenging space GNC scenario. The consortium explores the validation of AI-based ML techniques for robustness (maintaining performance over the field of use) and explainability (justification of actions), to increase the level of trust for space engineers to adopt those schemes.
In this project, Numalis and LNE decided to implement a neural network validation strategy adapted for optronic semantic analysis. In other words, the validation of neural networks for satellite image segmentation. Together, Numalis and LNE have defined an experimental framework using formal methods to extract and present objective information on the behavior of the network and, more specifically, the relationship between its robustness (resistance to noise) and its performance (ability to classify correctly).
Collaborative project between Synapse défense, Delfox and Numalis, on a subject provided by Dassault Aviation. The aim was to develop an AI algorithm for simulating dogfight (air combat) and to explain the AI behavior.
Numalis was involved in creating the AI training platform using reinforcement learning and its tools were used to provide relevance to determine the perceptions used by neural networks for navigation or fire control choices. Numalis has also defined a methodology for providing semantic feedback from relevance results to explain behavior to a human agent in an understandable way.
This project aimed to define the first tools using formal methods for neural network analysis. Funded by the DGA (a department of the French Ministry of the Armed Forces whose mission is to prepare the future of French defense systems), this programme also brought together a group of industrial experts including ADS, Thales, Thales Alenia Space, MBDA, Safran, Naval Group, Continental, Renault and C-S.
Contact us today to learn more about how Numalis can help you confidently adopt and deploy AI
