INRIA a CCSD electronic archive server
Not a member yet
122212 research outputs found
Sort by
Delayed High Order Sliding Mode Control Using Implicit Lyapunov Function Approach
No full textInternational audienceThis paper presents a novel delayed high-order sliding mode control strategy supported by dedicated mathematical tools. Building on the implicit Lyapunov-Razumikhin function method for establishing accelerated (hyperexponential) stability in time-delay systems and the design of high-order sliding mode controls for chain of integrators, we propose a time-delay modification/approximation of a sliding mode feedback, preserving the key advantages of this approach, such as the ability to reject matched perturbations and hyperexponential convergence, while significantly reducing chattering. We validate our theoretical findings through numerical simulations, providing empirical support for the effectiveness of the proposed method
A toy model for frequency cascade in the nonlinear Schrodinger equation
No full text24 pagesWe present an elementary approach to observe frequency cascade on forced nonlinear Schrödinger equations. The forcing term consists of a constant term, perturbed by a modulated Gaussian well. Algebraic computations provide an explicit frequency cascade when time and space derivatives are discarded from the nonlinear Schrödinger equation. We provide stability results, showing that when derivatives are incorporated in the model, the initial algebraic solution may be little affected, possibly over long time intervals. Numerical simulations are provided, which support the analysis
Automatic Lighthouse Calibration Using Conics for Indoor Robot Localization
No full textInternational audienceIn this letter, we propose a technique for calibrating Lighthouse localization systems using a single view of two or more coplanar circles traced by a moving robot. The calibration method leverages conic algebra to compute the homography between the Lighthouse view and the world plane, up to similarity. This approach requires minimal user intervention and is particularly suited for automatically calibrating large-scale deployments involving hundreds of mobile robots. We validate our method using a centimeter-scale differential drive robot, utilizing 5 cm circles to calibrate a 2×2m 2 area. The proposed technique achieved a mean positional accuracy of 7.77 mm, compared to the 5.37 mm accuracy of a previous calibration method based on manual measurements and known correspondences. We demonstrate that the conics traced by the robot are accurate enough for reliable homography estimation, even under varying conditions of tire material and surface type. A camera-based motion capture system served as the ground truth for all experiments. This work represents a step toward scalable and decentralized lighthouse calibration, enabling efficient 2D localization in large-scale robotic systems
Searching for Attractors : To infinity and beyond
No full textInternational audienceDiscrete and nondeterministic modelling of regulatory and signalling networks allows to characterize many of their crucial dynamical properties, with a reasonable computational effort. A dynamical feature of particular interest in its own right, as well as in terms of biological relevance, is the landscape of attractors and their attraction basins. In recent years, we have developped new approaches to discovery and global cartography of this basin landscape. We enlarge the set of models used, by moving to Petri nets on the one hand and an opening to continuous dynamics on the other. Surprisingly, these openings are rewarded not only by a sharpening of the analysis, but also the emergence of compact and readable data structures, and fast search algorithms. The talk will advocate the cross-fertilization between discrete and continuous approaches, and to not be afraid of pushing limits.</div
Disentangling Myth from Reality: A Plea for exploring New Avenues
No full textInternational audienceDemonstrating the paralogism at the root of so-called nonlocality in QM. Proponents of nonlocality or many-worlds don't use Bayesian probability nor classical mechanics appropriately.Further pleading for a shift in viewpoint.Author G.A. Main claim from joint work with M. K
Median nerve stimulation to predict MI-BCI performances
No full textInternational audienceApproximately 30% of individuals fail to effectively use a Brain-Computer Interface (BCI), a phenomenon known as BCI deficiency [1]. Predicting BCI performance is thus crucial for optimizing system parameters, selecting users, and harmonizing participant groups. While BCI performance prediction based on motor imagery (MI) remains an open question, various neurophysiological predictors assess motor cortex activation ability [2, 3, 4, 5]. We propose a novel predictor based on Median Nerve Stimulation (MNS) [6, 7], specifically, the minimum value (200–800 ms post-MNS) of the Event-Related Desynchronization (ERD) at electrode C3 using a small Laplacian filter. Right-hand MI vs. rest BCI performance was evaluated offline using a Tangent Space Logistic Regression classifier in 31 subjects. BCI accuracy strongly correlated with post-MNS ERD (Spearman’s rho = -0.71, p < 0.001) [8]. Beyond correlation analysis, we actually predicted BCI performance using a Least Absolute Shrinkage and Selection Operator (LASSO) regressionmodel, trained on six MNS-based features: minimum ERD (200–800 ms) and maximum ERS (800–1500 ms) post-MNS in mu, beta, and mu+beta. Using only these features, LASSO predicted MI-BCI accuracies with a correlation of rho = 0.65 (p < 0.01) between real and predicted accuracies (Fig. 1A). We also tested whether the three post-MNS ERD could predict a performance group, rather than the exact accuracy score (Fig. 1B) [9]. LASSO achieved 74.19% accuracy for two groups, though performance decreased to 45.16% for three groups . Based on reports from the literature, our new MNS based predictor seems to outperform state-of-the-art alternatives, including SMR and MeanSP (rho = 0.53) [2, 5], PPfactor (rho = 0.48) [4], and Spectral Entropy (rho = 0.65) [3]. These results suggest an inherent neurophysiological predisposition for MI-BCI success. Future work will integrate multiple predictors into a single model for improved accuracy
Comparison of machine learning and human prediction to identify trauma patients in need of hemorrhage control resuscitation (ShockMatrix study): a prospective observational study
No full textInternational audienceBackgroundMachine learning could improve the timely identification of trauma patients in need of hemorrhage control resuscitation (HCR), but the real-life performance remains unknown. The ShockMatrix study aimed to compare the predictive performance of a machine learning algorithm with that of clinicians in identifying the need for HCR.MethodsProspective, observational study in eight level-1 trauma centers. Upon receiving a prealert call, trauma clinicians in the resuscitation room entered nine predictor variables into a dedicated smartphone app and provided a subjective prediction of the need for HCR. These predictors matched those used in the machine learning model. The primary outcome, need for HCR, was defined as: transfusion in the resuscitation room, transfusion of more than four red blood cell units in 6 h of admission, any hemorrhage control procedure within 6 h, or death from hemorrhage within 24 h. The human and machine learning performances were assessed by sensitivity, specificity, positive likelihood ratio, negative likelihood ratio, and net clinical benefit. Human and machine learning agreement was assessed with Cohen's kappa coefficient.FindingsBetween August 2022 and June 2024, out of 5550 potential eligible patients, 1292 were ultimately included in the analyses. The need for HCR occurred in 170/1292 patients (13%). The results showed a positive likelihood ratio of 3.74 (95% confidence interval [CI]: 3.20–4.36) and a negative likelihood ratio of 0.36 (95% CI: 0.29–0.46) for the human prediction and a positive likelihood ratio of 4.01 (95% CI: 3.43–4.70) and negative likelihood ratio of 0.35 (95% CI: 0.38–0.44) for the machine learning prediction. The combined use of human and machine learning prediction yielded a sensitivity of 83% (95% CI: 77–88%) and a specificity of 73% (95% CI: 70–75%). The Cohen's kappa coefficient showed an agreement of 0.51 (95% CI: 0.48–0.55).InterpretationThe prospective ShockMatrix temporal validation study suggests a comparable human and machine learning performance to predict the need for HCR using real-life and real-time information with a moderate level of agreement between the two. Machine learning enhanced decision awareness could potentially improve the detection of patients in need of HCR if used by clinicians
The Fair Game: Auditing & debiasing AI algorithms over time
No full textInternational audienceAbstract An emerging field of AI, namely Fair Machine Learning (ML), aims to quantify different types of bias (also known as unfairness) exhibited in the predictions of ML algorithms, and to design new algorithms to mitigate them. Often, the definitions of bias used in the literature are observational, i.e. they use the input and output of a pre-trained algorithm to quantify a bias under concern. In reality, these definitions are often conflicting in nature and can only be deployed if either the ground truth is known or only in retrospect after deploying the algorithm. Thus, there is a gap between what we want Fair ML to achieve and what it does in a dynamic social environment. Hence, we propose an alternative dynamic mechanism, “Fair Game”, to assure fairness in the predictions of an ML algorithm and to adapt its predictions as the society interacts with the algorithm over time. “Fair Game” puts together an Auditor and a Debiasing algorithm in a loop around an ML algorithm. The “Fair Game” puts these two components in a loop by leveraging Reinforcement Learning (RL). RL algorithms interact with an environment to take decisions, which yields new observations (also known as data/feedback) from the environment and in turn, adapts future decisions. RL is already used in algorithms with pre-fixed long-term fairness goals. “Fair Game” provides a unique framework where the fairness goals can be adapted over time by only modifying the auditor and the different biases it quantifies. Thus, “Fair Game” aims to simulate the evolution of ethical and legal frameworks in the society by creating an auditor which sends feedback to a debiasing algorithm deployed around an ML system. This allows us to develop a flexible and adaptive-over-time framework to build Fair ML systems pre- and post-deployment
FADE: federated aggregation with discrimination elimination
No full textInternational audienceIn this work, we investigate how unfair updates with opposing biases can cancel each other out during aggregation in federated learning (FL), leading to a fairer overall model from a group fairness perspective. We analytically and empirically analyze this Federated Aggregation with Discrimination Elimination (FADE) phenomenon, considering both linear and nonlinear models. In addition, we build on this observation and introduce two novel fairness-aware FL aggregation strategies. The first strategy, FADE-OptW, uses sequential optimization to optimize weights assigned to each client based on their fairness levels. The second approach, FADE-SSP, identifies the optimal subset of clients that minimizes the weighted average fairness level at each round along the convergence path, and for a given metric. Our experiments demonstrate significant improvements in fairness, achieving up to a 60% reduction in discrimination compared to standard FedAvg-based FL. We achieve these gains while maintaining the model's predictive performance on highly heterogeneous client data distributions
Optimisation de réseaux de neurones : algorithmes et logiciel pour un système électrique durable
No full textCurrent technologies only allow storage by expensive and inefficient means, which makes it difficult to store electricity on a large scale. For the grid to function properly, electricity fed into the grid must match electricity used at all times. Historically, and still today, production resources are planned in advance of demand to maintain this balance. It is therefore crucial to forecast electricity consumption as accurately as possible. The integration of renewable energies, whose production is intermittent and dependent on weather conditions, is making the balance increasingly unstable. Managing this is becoming more complex, making forecasting wind and photovoltaic production now essential.Statistical learning models are used to make consumption and production forecasts. These models take past values and data from explanatory variables and use them to model the signal. To build efficient models, one must choose the input variables, the type of model, and its parameters. Given the vast number of signals to be forecasted, it would be beneficial to automate these choices to create competitive models. Automated Machine Learning (AutoML) is the process of automating the generation of learning models optimized according to the use case. Over the last ten years, numerous AutoML tools have been developed. However, most of them focus on optimizing classification or regression models on tabular data, or on optimizing neural network architectures for image or text processing. These tools are not appropriate for optimizing electricity consumption and production forecasting models.This thesis is a progress towards automating the generation of time series forecasting models required for power system management. The research work focused on developing the DRAGON Python package, which offers a range of tools for specific yet widely used models: neural networks. DRAGON can be used to create flexible search spaces encompassing a wide variety of neural networks by simultaneity optimizing the architecture and the hyperparameters. They are encoded by Directed Acyclic Graphs (DAGs), where the nodes are operations, parameterised by various hyperparameters, and the edges are the connections between these nodes. To navigate these graph-based search spaces and optimize their structures, the package proposes various search algorithms based on meta-heuristics and bandits-approaches. This thesis details how DRAGON is used for electricity consumption and production forecasts, enabling state-of-the-art models to be generated for these two industrial use cases.Les technologies actuelles ne permettant le stockage que par des moyens coûteux et peu efficaces, l'électricité reste difficile à stocker à grande échelle. Pour le bon fonctionnement du réseau, il est ainsi important qu'à tout instant, l'électricité injectée dans le réseau soit égale à l'électricité consommée. Historiquement et encore aujourd'hui, pour maintenir cet équilibre, les moyens de production sont planifiés par anticipation de la demande; d'où l'importance de prévoir aussi précisément que possible la consommation électrique. Avec l'intégration massive des énergies renouvelables dont la production est intermittente et dépendante des conditions météorologiques, la production devient de plus en plus instable et la gestion de l'équilibre se complexifie : des prévisions des productions éolienne et photovoltaïque sont désormais indispensables.Les prévisions de consommation et de production sont réalisées à l'aide de modèles d'apprentissage statistique, qui modélisent le signal en se basant sur ses valeurs passées et des données de variables dites explicatives. Pour construire un modèle performant, il est nécessaire de choisir les variables explicatives considérées, le type de modèle ainsi que sa paramétrisation. Au vu du très grand nombre de signaux à prévoir, il pourrait être intéressant d'automatiser ces choix pour créer automatiquement des modèles compétitifs. Le textit{Machine Learning} automatisé, également appelé AutoML pour textit{Automated Machine Learning}, est le processus d'automatisation de la génération de modèles d'apprentissage optimisés en fonction du cas d'usage. De nombreux outils d'AutoML ont été développés depuis une dizaine d'années, mais la plupart se concentrent sur l'optimisation de modèles de classification ou de régression sur des données tabulaires, ou sur l'optimisation d'architectures de réseaux de neurones pour le traitement d'images ou de textes. Ils ne sont donc pas forcément adaptés à la prévision de séries temporelles telles que la consommation ou production électrique.Cette thèse est un premier pas vers l'automatisation de la génération de modèles pour les prévisions des séries temporelles nécessaires à la gestion du système électrique. Les travaux de recherche se sont concentrés sur le développement du textit{package} Python DRAGON, qui propose divers outils pour optimiser des modèles bien particuliers, mais largement utilisés: les réseaux de neurones. Le textit{package} rend possible la création d'espaces de recherche plus ou moins flexibles, englobant une grande diversité d'architectures et qui permettent d'optimiser à la fois l'architecture et les hyperparamètres. Ces espaces de recherche sont encodés par des graphes acycliques dirigés, où les nœuds sont des opérations, paramétrées par divers hyperparamètres, et les arêtes sont les connexions entre ces nœuds. Afin de naviguer dans ces espaces de recherche à base de graphes et d'en optimiser les structures, divers algorithmes de recherche à base de métaheuristiques et de bandits sont proposés dans le textit{package}. Après une présentation de DRAGON, cette thèse détaille comment ce textit{package} est utilisé pour les prévisions de consommation et de production électrique et permet de générer des modèles à l'état de l'art dans ces deux cas d'usage industriels