Archive ouverte de l'ENAC
Not a member yet
3458 research outputs found
Sort by
Rigorous Floating-Point to Fixed-Point Quantization of Deep Neural Networks on STM32 Micro-controllers
No full textInternational audienceEmbedding artifcial intelligence onto lowpower devices is a challenging task that has been partiallyovercome by recent advances in machine learning andhardware design. Currently, deep neural networks can bedeployed on embedded targets to perform various taskssuch as speech recognition, object detection or humanactivity recognition. However, it is still possible to optimizedeep neural networks on embedded devices. These optimizations mainly concern energy consumption, memoryand real-time constraints, but also easier deployment atthe edge. In addition, there is still a need for a betterunderstanding of what can be achieved for different usecases. This work focuses on the quantization and deployment of deep neural networks on low-power 32-bit microcontrollers. In this article, the quantization method used isbased on solving an integer optimization problem derivedfrom the neural network model and concerning the accuracy of the computations and results at each point of thenetwork. We evaluate the performance of our quantizationmethod on a collection of neural networks measuring theanalysis time and time-to-solution improvement betweenthe foating- and fxed-point networks, considering a typicalembedded platform employing a STM32 Nucleo-144 microcontroller
Propagation électromagnétique en atmosphère turbulente inhomogène réaliste
No full textInternational audienceLa stochastic multiple phase screen method(sMPS) est largement utilisée pour modéliser l’impact dela turbulence troposphérique sur la propagation d’ondesélectromagnétiques. Elle repose généralement sur l’utili-sation de spectres de Kolmogorov homogènes. Cette étudeprésente deux méthodes pour générer une turbulence in-homogène avec la sMPS. L’approche LES-Kolmogorovutilise des profils verticaux de la constante de structureturbulente directement extraits de simulation atmosphé-riques, tandis que la méthode Bump-Kolmogorov (BK-model) les reproduit approximativement avec un modèleparamétrique. Les calculs de profils de log-amplitudemettent en évidence l’impact spécifique de l’inhomogé-néité de la turbulence ainsi qu’une bonne concordance desdeux méthodes
Challenges of Music Score Writing and the Potentials of Interactive Surfaces
No full textInternational audienceComposers use music notation programs throughout their creative process. Those programs are essentially elaborate structured document editors that enable composers to create high-quality scores by enforcing musical notation rules. They effectively support music engraving, but impede the more creative stages in the composition process because of their lack of flexibility. Composers thus often combine these desktop tools with other mediums such as paper. Interactive surfaces that support pen and touch input have the potential to address the tension between the contradicting needs for structure and flexibility. We interview nine professional composers. We report insights about their thought process and creative intentions, and rely on the ``Cognitive Dimensions of Notations'' framework to capture the frictions they experience when materializing those intentions on a score. We then discuss how interactive surfaces could increase flexibility by temporarily breaking the structure when manipulating the notation
MIQCQP reformulation of the ReLU neural networks Lipschitz constant estimation problem
No full textIt is well established that to ensure or certify the robustness of a neural network, its Lipschitz constant plays a prominent role. However, its calculation is NP-hard. In this note, by taking into account activation regions at each layer as new constraints, we propose new quadratically constrained MIP formulations for the neural network Lipschitz estimation problem. The solutions of these problems give lower bounds and upper bounds of the Lipschitz constant and we detail conditions when they coincide with the exact Lipschitz constant
The robust cyclic job shop problem
No full textInternational audienceThis paper deals with the cyclic job shop problem where the task durations are uncertain and belong to a polyhedral uncertainty set. We formulate the cyclic job shop problem as a two-stage robust optimization model. The cycle time and the execution order of tasks executed on the same machines correspond to the here-and-now decisions and have to be decided before the realization of the uncertainty. The starting times of tasks corresponding to the wait-and-see decisions are delayed and can be adjusted after the uncertain parameters are known. In the last decades, different solution approaches have been developed for two-stage robust optimization problems. Among them, the use of affine policies, row and row-and-column generation algorithms are the most common. In this paper, we propose a branch-andbound algorithm to tackle the robust cyclic job shop problem with cycle time minimization. The algorithm uses, at each node of the search tree, a robust version of the Howard's algorithm to derive a lower bound on the optimal cycle time. Moreover, we design a row generation algorithm and a column-and-row generation algorithm and compare it to the branchand-bound method. Finally, encouraging preliminary results on numerical experiments performed on randomly generated instances are presented
A time-dependent subgraph-capacity model for multiple shortest paths and application to CO2/contrail-safe aircraft trajectories
No full textInternational audienceThis paper proposes a study motivated by the problem of minimizing the environmental impact of air transport at the level of a complete air network, considering thereby several aircraft. Both CO2 and non-CO2 effects are taken into account to calculate this impact. The proposed methodology takes into account a network point of view where airspace capacities evolve as well as the traffic itself over time. Finding a shortest path with various constraints and cost functions is a common problem in operations research. This particular study deals with the special case of computing multiple shortest paths with capacity constraints on a time-dependent subgraph. Multiple shortest paths is understood as one shortest path for each vehicle considered. The static special case is modeled as a Mixed Integer Linear Program (MILP), so that it can be solved directly by a standard commercial solver. The time-dependent nature of the problem is then modeled thanks to a sliding-window approach. Encouraging numerical results on the CO2/contrail-safe aircraft trajectories application are obtained and show that the environmental impact can be significantly reduced while maintaining safety by satisfying the airspace capacity constraints
The Impact of Age on Cognitive Performance at Work: A Focus on Pilots and Air Traffic Controllers
No full textInternational audienceThe aging of the working population is the natural consequence of the demographic evolution of western countries. Managing this change is a real challenge for society and it raises the issue of the retirement age extension. Beyond economic considerations, it is imperative to explore strategies for preserving performance and safety at work with older individuals. This is naturally true for physical jobs, but also for critical and cognitively demanding fields of activity, involving a rapid pace of decision-making such as piloting (van Drongelen et al., 2017) and air traffic control (ATC) (Heslegrave, 1998). In 2023, the CEO of the Regional Airline Association stated that approximately 50% of the commercial airline workforce will retire within the next 15 years. The shortage of aviation professionals is thus a major issue for local and national authorities and companies in the coming years. This retirement trend is anticipated also for air traffic controllers (ATCO), although the specific numbers may vary from country to country. In this chapter we will describe the effects of aging on cognitive performance of aviation professionals, and the way in which the authorities regulate its potential consequences for safety. We then present the biological basis of age effects, with an emphasis on the CRUNCH (Compensation-Related Utilisation of Neural Circuits Hypothesis ) model. A number of studies on aging in the fields of piloting and air ATC are presented. Finally, we outline some prospects for objective measurements of aging using physiological measures
A Catalogue of Deconfliction Actions Extracted from Historical ADS-B Data
No full textInternational audienceA conflict in Air Traffic Management is defined by a potential future risk of loss of separation. To solve a conflict, air traffic controllers take proactive actions to ensure safe separations between aircraft. They issue specific instructions to pilots for corrective measures, such as lateral manoeuvres, changes in altitude or speed adjustments. To alleviate the workload of controllers, the conflict detection and resolution process can be automated, resulting in recommendations for efficient manoeuvres. Traditional conflict resolution algorithms often neglect factors inherent to controllers' decision-making, leading to seemingly impractical manoeuvre suggestions from a human standpoint, causing reluctance in acceptation among controllers. The aim of our research is to obtain a catalogue of prevalent deconfliction practices, incorporating controllers' uncertainty models derived from actual flight data. In the current contribution, we focus on lateral deconfliction manoeuvres in en-route air traffic, and implement a simple heuristic method to extract a catalogue of resolved conflict situations from historical ADS-B data. This catalogue will provide insights into controllers' decision-making processes. In future works, we intend to use this catalogue to identify the best practices for traffic deconfliction, taking into account human factors and operational uncertainties, and to incorporate them into conflict resolution algorithms.</div
The UAV Feasibility Trajectory Prediction Using Convolution Neural Networks: Wind direction and uncertainty are crucial in aircraft or unmanned aerial vehicle trajectories. By computing wind covariance matrices on each spatial grid point, these spatial grids can be defined as images with symmetric positive definite matrix elements. A data pre-processing step, a specific convolution, a specific max-pooling, and specific flatten layers are implemented to process such images. Then, the neural network is applied to spatial grids, whose elements are wind covariance matrices, to solve classification problems related to the feasibility of unmanned aerial vehicles based on wind direction and wind uncertainty.
No full textInternational audienceWind direction and uncertainty are crucial in aircraft or unmanned aerial vehicle trajectories. By computing wind covariance matrices on each spatial grid point, these spatial grids can be defined as images with symmetric positive definite matrix elements. A data pre-processing step, a specific convolution, a specific max-pooling, and specific flatten layers are implemented to process such images. Then, the neural network is applied to spatial grids, whose elements are wind covariance matrices, to solve classification problems related to the feasibility of unmanned aerial vehicles based on wind direction and wind uncertainty
Assessing jamming and spoofing impacts on GNSS receivers: Automatic gain control (AGC)
No full textInternational audienceIn modern GNSS receivers, the Automatic Gain Control (AGC) monitors the received signal level to optimize quantization and mitigate interference. This paper characterizes the jamming and spoofing impact on AGC and received signal. It first expresses the AGC gain as a function of the received signal level. Under nominal conditions, the AGC leverages the ergodic properties of the received signal to estimate its level over time. Two physical quantities, namely time-based power and signal distribution, are typically considered. However, in the presence of interference, these ergodic properties are no longer guaranteed, posing challenges in modeling the behavior of these quantities. This paper proposes a probabilistic framework for interpreting temporal estimation and computing time-based power and distribution in order to characterize AGC gain under jamming and spoofing. First, this study models the spoofing impact for both unique and multiple emitted spoofing signals as a function of the re-radiated noise power and the spoofing signals' characteristics (e.g., number of emitted signals, amplitudes, modulation). Furthermore, it reveals the non-uniformity of jamming chirp phase, which introduces distortions in power and signal distribution, consequently affecting AGC gain, and demonstrates the convergence of the jamming signal toward a continuous wave signal at high frequencies.</div