Portail HAL Ensta
Not a member yet
11080 research outputs found
Sort by
The SocialAI school: a framework leveraging developmental psychology toward artificial socio-cultural agents
No full textInternational audienceDevelopmental psychologists have long-established socio-cognitive abilities as fundamental to human intelligence and development. These abilities enable individuals to enter, learn from, and contribute to a surrounding culture. This drives the process of cumulative cultural evolution, which is responsible for humanity's most remarkable achievements. AI research on social interactive agents mostly concerns the emergence of culture in a multi-agent setting (often without a strong grounding in developmental psychology). We argue that AI research should be informed by psychology and study socio-cognitive abilities enabling to enter a culture as well. We draw inspiration from the work of Michael Tomasello and Jerome Bruner, who studied socio-cognitive development and emphasized the influence of a cultural environment on intelligence. We outline a broader set of concepts than those currently studied in AI to provide a foundation for research in artificial social intelligence. Those concepts include social cognition (joint attention, perspective taking), communication, social learning, formats, and scaffolding. To facilitate research in this domain, we present The SocialAI school—a tool that offers a customizable parameterized suite of procedurally generated environments. This tool simplifies experimentation with the introduced concepts. Additionally, these environments can be used both with multimodal RL agents, or with pure-text Large Language Models (LLMs) as interactive agents. Through a series of case studies, we demonstrate the versatility of the SocialAI school for studying both RL and LLM-based agents. Our motivation is to engage the AI community around social intelligence informed by developmental psychology, and to provide a user-friendly resource and tool for initial investigations in this direction. Refer to the project website for code and additional resources: https://sites.google.com/view/socialai-school
Un nouveau regard sur la crise de productivité des Big Pharma
No full textInternational audienceDepuis de nombreuses années, les grandes entreprises historiques de l’industrie biopharmaceutique (Big Pharma) font face à une « crise de productivité ». Elle se traduit par un nombre stagnant voire décroissant de mises sur le marché de nouveaux médicaments malgré une hausse importante des dépenses de recherche et développement (R&D). Les enjeux liés à cette crise vont au-delà de ces dynamiques industrielles constatées. Ils ont trait également à une crise de la mesure de l’inventivité. Dans cet article, nous intégrons, dans la mesure de l’inventivité fondée sur les bases de connaissances des entreprises, les usages des connaissances créées. En économie de l’innovation, la « qualité » d’une connaissance intègre en effet cette dimension, celle de servir de source à des créations de connaissances futures. Nous évaluons également empiriquement le lien entre inventivité et taille de l’entreprise dans la biopharmacie, taille envisagée par le montant des dépenses en R&D mais aussi en termes de nombre d’employés et le chiffre d'affaires engendré. Nous aboutissons alors à un constat qui nous ramène à la « crise » évoquée p^lus haut : l’inventivité est en relation inverse avec les montants de R&D dépensés par les entreprises du secteur
Collective Innovation in Groups of Large Language Models
No full textInternational audienceHuman culture relies on collective innovation: our ability to continuously explore how existing elements in our environment can be combined to create new ones. Language is hypothesized to play a key role in human culture, driving individual cognitive capacities and shaping communication. Yet the majority of models of collective innovation assign no cognitive capacities or language abilities to agents. Here, we contribute a computational study of collective innovation where agents are Large Language Models (LLMs) that play Little Alchemy 2, a creative video game originally developed for humans that, as we argue, captures useful aspects of innovation landscapes not present in previous test-beds. We, first, study an LLM in isolation and discover that it exhibits both useful skills and crucial limitations. We, then, study groups of LLMs that share information related to their behaviour and focus on the effect of social connectivity on collective performance. In agreement with previous human and computational studies, we observe that groups with dynamic connectivity out-compete fully-connected groups. Our work reveals opportunities and challenges for future studies of collective innovation that are becoming increasingly relevant as Generative Artificial Intelligence algorithms and humans innovate alongside each other
A Combined CNN-LSTM Network for Ship Classification on SAR Images
No full textInternational audienceSatellite SAR (synthetic aperture radar) imagery offers global coverage and all-weather recording capabilities, making it valuable for applications like remote sensing and maritime surveillance. However, its use in machine learning-based automatic target classification faces challenges, including the limited availability of SAR target training samples and the inherent constraints of SAR images, which provide less detailed features compared to natural images. These issues hinder the effective training of convolutional neural networks (CNNs) and complicate the transfer learning process due to the distinct imaging mechanisms of SAR and natural images. To address these challenges, we propose a shallow CNN architecture specifically designed to optimize performance on SAR datasets. Evaluations were performed on three datasets: FUSAR-Ship, OpenSARShip, and MSTAR. While the FUSAR-Ship and OpenSARShip datasets present difficulties due to their limited and imbalanced class distributions, MSTAR serves as a benchmark with balanced classes. To compare and optimize the proposed shallow architecture, we examine various properties of CNN components, such as the filter numbers and sizes in the convolution layers, to reduce redundancy, improve discrimination capability, and decrease network size and learning time. In the second phase of this paper, we combine the CNN with Long short-term memory (LSTM) networks to enhance SAR image classification. Comparative experiments with six state-of-the-art CNN architectures (VGG16, ResNet50, Xception, DenseNet121, EfficientNetB0, and MobileNetV2) demonstrate the superiority of the proposed approach, achieving competitive accuracy while significantly reducing training times and network complexity. This study underscores the potential of customized architectures to address SAR-specific challenges and enhance the efficiency of target classification
Guided Drawing in Medical Education: Evaluating AR and MR Systems for Anatomy Learning
No full textInternational audienc
Passive acoustic monitoring for assessing marine mammals population in European waters: Workshop conclusions and perspectives
No full textInternational audienc
Extensions and Scalability Experiments of a Generic Model-Driven Architecture for Variability Model Reasoning
No full textInternational audienceUntil recently, the state-of-the-art of Software Product Line (SPL) configuration and verification automation consisted of a collection of ad-hoc approaches tightly coupling a single input Variability Modeling Language (VML) with a single constraint solver. To remedy this situation, a novel generic model-driven architecture was then proposed that enables using a variety of VMLs and solvers. The key ideas of this proposal were (a) the use of a standard logical language (CLIF) as a pivot between VMLs and solvers, and (b) the use of a standard data exchange format (JSON) to explicilty and declaratively specify the abstract syntax and semantics of the VMLs to be used in an SPL engineering project and the automated reasoning task to be performed by the solvers.In this article, we overcome the limitations of this initial proposal in three key ways: (1) we add the ability to reason on textual or hybrid VMLs, rather than only on diagrammatic VMLs, enhancing the versatility of the architecture on the input side; (2) we enable the use of solvers from a third paradigm, enhancing the versatility of the architecture on the output side; and, (3) we present the results of scalability performance experiments of an implementation of this architecture. These results have been achieved without significantly altering the architecture, demonstrating its agnosticism with respect to specific VMLs and solvers. It also shows that it can underlie the implementation of practical variability reasoning tools that scale up to real sized variability model analysis and configuration needs.</p
Cycloadditions of 5-Vinyloxazolidine-2,4-diones: A Straightfor-ward Access to the (thio)Hydantoin Scaffold
No full textA palladium-catalyzed (3+2) cycloaddition between 5-vinyloxazolidine-2,4-diones (VOxD) and (thio)isocyanates is described. Under optimized conditions, an array of (thio)hydantoins was readily prepared and an enantioselective version of this transformation was then studied. To illustrate the importance of this method, a concise synthesis of two bioactive compounds, nirvanol and mephenytoin, was carried out. This work emphasizes the synthetic potential of VOxD as useful precursors of zwitterionic aza-π- allylpalladium II intermediates
Path Planning for Unmanned Aerial Vehicles in Dynamic Environments: A Novel Approach Using Improved A * and Grey Wolf Optimizer
No full textInternational audienceUnmanned aerial vehicles (UAVs) play pivotal roles in various applications, from surveillance to delivery services. Efficient path planning for UAVs in dynamic environments with obstacles and moving landing stations is essential to ensure safe and reliable operations. In this study, we propose a novel approach that combines the A* algorithm with the grey wolf optimizer (GWO) for path planning, referred to as GW-A*. Our approach enhances the traditional A algorithm by incorporating weighted nodes, where the weights are determined based on the distance from obstacles and further optimized using GWO. A simulation using dynamic factors such as wind direction and wind speed, which affect the quadrotor UAV in the presence of obstacles, was used to test the new approach, and we compared it with the A* algorithm using various heuristics. The results showed that GW-A* outperformed A* in most scenarios with high and low wind speeds, offering more efficient paths and greater adaptability
Fonctionnalisations régiosélectives des naphtalènes via des activations C-H : nouvelles méthodologies et applications en synthèse
No full textThis thesis deals with regioselective C-H activation reactions on naphthalene. The objective is to develop several methodologies that can be applied in the synthesis of products of interest.The first project of this thesis enabled the functionalization of 1-carbonyl-naphthalenes by performing fluoroalkylations and fluoroalkenylations at the C8 position through C-H activation, catalyzed by palladium. Various directing groups were employed, and it was found that the methylene amide and methylene ketone are the best candidates for this type of reactivity. The potential of these reactions allowed the synthesis of numerous examples varied in terms of chemical functions and positioning on the naphthalene ring.The second project established the conditions for various halogenations of 1-naphthaldehyde through C-H activation, also catalyzed by palladium. We found conditions allowing for regioselective activation, either at C2 or C8. Although the scope of this method's tolerance is more limited than the previous one, it was possible to achieve various applications, enabling the synthesis of skeletons of products of interest.Finally, we applied the C8 halogenation method of 1-naphthaldehyde to the synthesis of a library of naphtholactams. These substrates were evaluated in biological tests against the bacterium Pseudomonas aeruginosa. In parallel, this method was employed in the synthesis of a natural product: aspergilline F, obtained in 7 steps from the corresponding 1-naphthaldehyde.Cette thèse traite des réactions d'activations C-H régiosélectives sur le naphtalène. L'objectif est de développer plusieurs méthodologies pouvant être impliquées en synthèse de produits d'intérêt.Le premier projet de cette thèse a permis de fonctionnaliser des 1-carbonyl-naphtalènes en réalisant des fluoroalkylations et des fluoroalcénylations en position C8 par activation C-H, catalysée par du palladium. Divers groupements directeurs ont été employés, et il a été mis en évidence que l'amide méthylique et la cétone méthylique sont les meilleurs candidats pour ce type de réactivité. Le potentiel de ces réactions a permis de synthétiser de nombreux exemples variés en termes de fonctions chimiques et de positionnement des fonctions évaluées sur le cycle du naphtalène.Le second projet a établi les conditions pour diverses halogénations du 1-naphtaldéhyde par activation C-H, également catalysées par du palladium. Nous avons trouvé des conditions permettant une activation régiosélective, soit en C2, soit en C8. Bien que l'étendue de la tolérance de cette méthode soit plus limitée que celle du premier projet, il a été possible de réaliser diverses applications, permettant la synthèse de squelettes de produits d'intérêt.Pour finir, nous avons appliqué la méthode d'halogénation en C8 du 1-naphtaldéhyde à la synthèse d'une librairie de naphtolactames. Ces substrats ont été évalués dans le cadre de tests biologiques contre la bactérie Pseudomonas aeruginosa. En parallèle, cette méthode a été employée dans la synthèse d'un produit naturel : l'aspergilline F, obtenue en 7 étapes à partir du 1-naphtaldéhyde correspondant