Search

Aim: Drug discovery is a long process, often taking decades of research endeavors. It is still an active area of research in both academic and industrial sectors with efforts on reducing time and cost. Computational simulations like molecular docking enable fast exploration of large databases of compounds and extract the most promising molecule candidates for further in vitro and in vivo tests. Structure-based molecular docking is a complex process mixing both surface exploration and energy estimation to find the minimal free energy of binding corresponding to the best interaction location. Methods: Hereafter, heterogeneous graph score (HGScore), a new scoring function is proposed and is developed in the context of a protein-small compound-complex. Each complex is represented by a heterogeneous graph allowing to separate edges according to their class (inter- or intra-molecular). Then a heterogeneous graph convolutional network (HGCN) is used allowing the discrimination of the information according to the edge crossed. In the end, the model produces the affinity score of the complex. Results: HGScore has been tested on the comparative assessment of scoring functions (CASF) 2013 and 2016 benchmarks for scoring, ranking, and docking powers. It has achieved good performances by outperforming classical methods and being among the best artificial intelligence (AI) methods. Conclusions: Thus, HGScore brings a new way to represent protein-ligand interactions. Using a representation that involves classical graph neural networks (GNNs) and splitting the learning process regarding the edge type makes the proposed model to be the best adapted for future transfer learning on other (protein-DNA, protein-sugar, protein-protein, etc.) biological complexes.

International audience; Artificial intelligence (AI) is often presented as a new Industrial Revolution. Many domains use AI, including molecular simulation for drug discovery. In this review, we provide an overview of ligand–protein molecular docking and how machine learning (ML), especially deep learning (DL), a subset of ML, is transforming the field by tackling the associated challenges.

International audience; With a turnover of more than 4.2 billion euros in 2020 and a 20% share in the value of the French wine industry's exports, the champagne industry represents a considerable weight in the French economy. In this context of significant economic development, the issue of climate change has been added, calling into question the practices and means of production of the sector. The challenges related to global warming and an ever-increasing demand for yield can be addressed using the Internet of Things (IoT) and Artificial Intelligence (AI) technologies to benefit champagne production and answer these challenges. This article presents a solution to optimise Vranken Pommery products' quality and make environmentally friendly decisions by using intelligent sensors distributed as close as possible to the production and storage facilities to collect data. These sensors use LoRaWAN technology and protocol to communicate. The system integrates components capable of hosting artificial intelligence algorithms and using advanced microcontrollers that allow for intelligent communication network implementation while reducing power consumption and deployment costs.

International audience; THIS IS A PREPRINT for Applied Intelligence. The revised version is available here: https://link.springer.com/article/10.1007/s10489-021-02359-6This paper examines multiple CNN-based (Convolutional Neural Network) models for Covid-19 forecast developed by our research team during the French lockdown. In an effort to understand and predict both the epidemic evolution and the impacts of this disease, we conceived models for multiple indicators: daily or cumulative confirmed cases, hospitalizations, hospitalizations with artificial ventilation, recoveries and deaths. In spite of the limited data available when the lockdown was declared, we achieved good short-term performances at the national level with a classical CNN for hospitalizations, leading to its integration into a hospitalizations surveillance tool after the lockdown ended. Also, A Temporal Convolutional Network with quantile regression was found successful at predicting multiple Covid-19 indicators at the national level by using data available at different scales (worldwide, national, regional). The accuracy of the regional predictions was improved by using a hierarchical pre-training scheme, and an efficient parallel implementation allows for quick training of multiple regional models. The resulting set of models represent a powerful tool for short-term Covid-19 forecasting at different geographical scales, complementing the toolboxes used by health organizations in France.

International audience; Post-harvest fruit grading is a necessary step to avoid disease related loss in quality. In this paper, a hierarchical method is proposed to (1) remove the background and (2) detect images that contains grape diseases(botrytis, oidium, acid rot). Satisfying segmentation performances were obtained by the proposed Lite Unet model with 92.9% IoU score and an average speed of 0.16s/image. A pretrained MobileNet-V2 model obtained 94% F1 score on disease classification. An optimized CNN reached a score of 89% with less than 10 times less parameters. The implementation of both segmentation and classification models on low-powered device would allow for real-time disease detection at the press.

This article investigates the problem of agricultural yield prediction, including that of soybeans. Soya is a very nutritious legume and one of the species producing the most protein per hectare. It is used in particular as a source of protein in animal feed and as an oilseed. However, agricultural systems are dependent on climatic variability, and farmers must deal with this factor to optimize their activities. This article describes a method for predicting soybean yields based on machine learning. The comparative study shows that one can obtain forecasts with less than 2% margin of error using the Random Forest algorithm. In addition, the results obtained in this study can be extended to many other crops such as maize or rice.

O presente trabalho objetivou identificar e analisar a dinâmica estratosferica dos eventos de influencia do Buraco de Ozonio Antartico (BOA) ocorridos sobre o Sul do Brasil no ano de 2016. Para atingir esse objetivo, buscou-se dias de reducao no conteudo de ozonio estratosferico com base em dados medios diarios da Coluna Total de Ozonio (CTO) obtidos pelo espectrofotometro Brewer instalado no Observatorio Espacial do Sul (OES/INPE) (29.443752 oS, 53.823084 oO; 488,7 metros) e pelo instrumento de medicao de Ozonio ( Ozone Monitoring Instrument - OMI) a bordo do satelite Aura da NASA. Para esses dias foram realizadas analises da dinâmica da estratosfera atraves de campos isentropicos de Vorticidade Potencial (PV) nos niveis de 600 e 700 K de temperatura potencial, utilizando as medias diarias de PV obtidas a partir das reanalises ERA-Interim do ECMWF ( European Centre for Medium-range Weather Forecast ) a fim de verificar a origem das massas de ar pobres em ozonio, e desta forma, a confirmacao da origem dessas massas atraves das trajetorias retroativas obtidas com uso do modelo HYSPLIT ( HYbrid Single-Particle Lagrangian Integrated Trajectory) da NOAA (National Oceanic and Atmospheric Administration ). Alem disso, imagens do conteudo de ozonio do satelite OMI foram utilizadas como tecnica complementar para verificar a atuacao do BOA sobre o Sul do Brasil. Usando a metodologia apresentada identificou-se 6 eventos de influencia do Buraco de Ozonio Antartico sobre o Sul do Brasil no ano de 2016 com percentagem de reducao media de 11% da CTO em relacao as climatologias mensais, confirmados pela passagem de filamentos polares estratosfericos sobre a regiao do OES, observados nos campos isentropicos de VP nos niveis de 600 e 700 K de temperatura potencial assim como pelas trajetorias retroativas de origem polar obtidas no modelo HYSPLIT.

International audience; Fog computing extends the Cloud Computing paradigm to the edge of the network, relying on computing services intelligently distributed to best meet the applications needs such as low communication latency, data caching or confidentiality reinforcement. While P2P is especially prone to implement Fog computing platforms, it usually lacks important elements such as controlling where the data is stored and who will handle the computing tasks. In this paper we propose both a mapping approach for data-locality and a location-aware scheduling for P2P-based middlewares, improving the data management performance on fog environments. Experimental results comparing the data access performances demonstrate the interest of such techniques.

International audience; Container-based virtualization has been investigated as an attractive solution to achieve isolation, flexibility and efficiency in a wide range of computational applications. In High Performance Computing, many applications rely on clusters to run multiple communicating processes using MPI (Message Passing Interface) communication protocol. Container clusters based on Docker Swarm or Kubernetes may bring benefits to HPC scenarios, but deploying MPI applications over such platforms is a challenging task. In this work, we propose a self-content Docker Swarm platform capable of supporting MPI applications, and validate it though the performance characterization of a meteorological scientific application.

This article proposes to improve Apache Hadoop scheduling through a context-aware approach. Apache Hadoop is the most popular implementation of the MapReduce paradigm for distributed computing, but its design does not adapt automatically to computing nodes’ context and capabilities. By introducing context-awareness into Hadoop, we intent to dynamically adapt its scheduling to the execution environment. This is a necessary feature in the context of pervasive grids, which are heterogeneous, dynamic and shared environments. The solution has been incorporated into Hadoop and assessed through controlled experiments. The experiments demonstrate that context-awareness provides comparative performance gains, especially when some of the resources disappear during execution.

International audience; A depletion event of 11% of the total ozone column over South Brazil was analyzed from observational data and numerical simulations. The meteorological fields resulting from the simulation did not show the presence of subtropical or polar jets over the study region. The real and simulated soundings showed the dryness at high and low tropospheric levels, pointing a stratospheric intrusion. The total ozone column values from ERA-5 reanalyses showed the maintenance of a region with low total ozone values in tropical latitudes and its advance over the study region, before and during the depletion event. The isentropic retro-trajectories of air parcels at different levels, obtained with the Hybrid Single Lagrangian Integrated Trajectory Model (HYSPLIT), confirmed the zonal transport from the tropical region over Southern Brazil. A combination of a cyclonic circulation at the 850 K stratospheric isentropic level and an intense anticyclonic circulation at the 440 K tropospheric level, over the depletion region, was identified as responsible for the tropopause lift and horizontal transport of ozone-rich air out of the column.

A framework to perform inverse docking was developed.Different strategies to distribute the docking procedure were implemented.Validation with experimental complex have been done.A docking test of one ligand versus 100 proteins was performed.A better conformational sampling is processed than current methods. Molecular docking is a widely used computational technique that allows studying structure-based interactions complexes between biological objects at the molecular scale. The purpose of the current work is to develop a set of tools that allows performing inverse docking, i.e., to test at a large scale a chemical ligand on a large dataset of proteins, which has several applications on the field of drug research. We developed different strategies to parallelize/distribute the docking procedure, as a way to efficiently exploit the computational performance of multi-core and multi-machine (cluster) environments. The experiments conducted to compare these different strategies encourage the search for decomposing strategies since it improves the execution of inverse docking.

Organizations need process data in real time. Cloud Computing is a common choice but it presents high latency on data transfer. As an alternative to reduces the latency, Edge computing based on System-on-a-Chip systems (SoC-based systems) can retain part of the data processing, allowing resource sharing among multiple requests through the use of OS-level virtualization. In this work, we analyze the performance of two popular OS-level virtualization solutions - Linux Containers (LXC) and Docker - for I/O-bound activities running in SoC systems, in order to understand how resource sharing affects the processing performance.

http://www.sciencedirect.com/science/article/pii/S1877050914006024; International audience; Data mining tools may be computationally demanding, so there is an increasing interest on parallel computing strategies to improve their performance. The popularization of Graphics Processing Units (GPUs) increased the computing power of current desktop computers, but desktop-based data mining tools do not usually take full advantage of these architectures. This paper exploits an approach to improve the performance of Weka, a popular data mining tool, through parallelization on GPU-accelerated machines. From the profiling of Weka object-oriented code, we chose to parallelize a matrix multiplication method using state-of-the-art tools. The implementation was merged into Weka so that we could analyze the impact of parallel execution on its performance. The results show a significant speedup on the target parallel architectures, compared to the original, sequential Weka code.

PurposeThe purpose of this paper is to explain one of the most important collective communication patterns used in scientific applications which is the complete exchange, also called All‐to‐All. Although efficient algorithms have been studied for specific networks, general solutions like those available in well‐known MPI distributions (e.g. the MPI_Alltoall operation) are strongly influenced by the congestion of network resources.Design/methodology/approachIn this paper we present an integrated approach to model the performance of the All‐to‐All collective operation, which consists in identifying a contention signature that characterizes a given network environment, using it to augment a contention‐free communication model.FindingsThis approach, assessed by experimental results, allows an accurate prediction of the performance of the All‐to‐All operation over different network architectures with a small overhead.Practical implicationsThe paper discusses the problem of network contention in a grid environment, studying some strategies to minimize the impact of contention on the performance of an All‐to‐All operation.Originality/valueThe approach used, assessed by experimental results, allows an accurate prediction of the performance of the All‐to‐All operation over different network architectures with a small overhead. Also discussed is the problem of network contention in a grid environment and some strategies to minimize the impact of contention on the performance of an All‐to‐All operation.

Today, IoT applications are heavily dependent on public cloud computing services to perform data storage and analysis. Unfortunately, the cloud computing paradigm is unable to meet the requirements of critical applications that require low latency or enhanced privacy levels. The deployment of private cloud services on top of pervasive grids represent an interesting alternative to traditional cloud infrastructures, allowing the use of near-environment resources for IoT data analysis tasks. In this work we discuss the challenges associated with the deployment of IoT services over pervasive environments, and present a study case deployed over CloudFIT, a computing middleware for pervasive systems. Hence, we evaluate the behavior of a data-intensive application under volatility and heterogeneity constraints, bringing to light to the use of low-end devices that are usually located at the proximity to IoT sensors/actuators.

IoT applications are the next important step towards the establishment of mobiquitous systems, but at the same time these environments raise important challenges when considering data distribution and processing. While most IoT applications today rely on clouds as back-end, critical applications that require fast response or enhanced privacy levels may require proximity services specially tailored to these needs. The deployment of private cloud services on top of pervasive grids represent an interesting alternative to traditional cloud infrastructures. In this work we present CloudFIT, a PaaS middleware that allows the creation of private clouds over pervasive environments. Using a Map-Reduce application as example, we show how CloudFIT provides both storage and data aggregation/analysis capabilities at the service of IoT networks.

Prototyping and testing distributed systems is considered to be a hard task because it is not always possible to reproduce a given sequence of events. While simulations may help on this task, they cannot replace test and validation with real systems. In this paper we present Docker-Hadoop, a container-based virtualization platform designed to prototype, test and deploy MapReduce applications and systems. This tool allowed us to test and reproduce fault-tolerance scenarios that are especially interesting in the context of the PER-MARE project, which aims at adapting the Hadoop framework to the case pervasive systems. Indeed, we developed a fault-tolerant component that can circumvent the limitations from original Hadoop and prevent the job scheduling stall in the case of failures or network disconnections. Thanks to Docker-Hadoop, we could easily prototype and test our improved Hadoop, with the first scalability and speedup results being presented in this paper.

Network security is in a daily evolving domain. Every day, new attacks, viruses, and intrusion techniques are released. Hence, network devices, enterprise servers, or personal computers are potential targets of these attacks. Current security solutions like firewalls, intrusion detection systems (IDS), and virtual private networks (VPN) are centralized solutions, which rely mostly on the analysis of inbound network connections. This approach notably forgets the effects of a rogue station, whose communications cannot be easily controlled unless the administrators establish a global authentication policy using methods like 802.1x to control all network communications among each device. To the best of the authors’ knowledge, a distributed and easily manageable solution for the global security of an enterprise network does not exist. In this chapter, they present a new approach to deploy a distributed security solution where communication between each device can be control in a collaborative manner. Indeed, each device has its own security rules, which can be shared and improved through exchanges with others devices. With this new approach, called grid of security, a community of devices ensures that a device is trustworthy and that communications between devices progress in respect of the control of the system policies. To support this approach, the authors present a new communication model that helps structuring the distribution of security services among the devices. This can secure both ad-hoc, local-area or enterprise networks in a decentralized manner, preventing the risk of a security breach in the case of a failure.

This paper deals with the problem of broadcasting for cluster of clusters The construction of partial minimum spanning trees being NP-complete, several heuristic algorithms have been already formulated Many of these heuristics (like the heuristic of Kruskal) use the shortest path to connect the components of the tree They are not relevant in case of forwarding or overlapping communication during a step of the algorithm In this paper, we study a new heuristic for the minimum broadcasting tree and we evaluate it through simulations with different communication parameters and also through real experimentation over the Grid'5000 testbed.

International audience; Recent advances in communication technology en- able the emergence of a new generation of applications that integrates mobile devices with classical high performance systems as part of a common computing environment. In such environ- ments, keeping the coherence of shared data (distributed objects, for example) represents a real challenge as communications are strongly influenced by the performance and the reliability of mobile devices (laptops, PDAs and cellular telephones) and wireless networks (WiFi, Bluetooth). Indeed, data incoherence may arise due to message losses or node volatility, which blocks the algorithms used to synchronize these data. In this paper, we analyze the main challenges concerning the manipulation of shared distributed objects in a pervasive environment. We demonstrate how a membership service can be enhanced to tolerate temporary disconnections and message losses without blocking, while reducing the number of exchanged message.

One of the most important collective communication patterns used in scientific applications is the complete exchange, also called All-to-All. Although efficient algorithms have been studied for specific networks, general solutions like those available in well-known MPI distributions (e.g. the MPI Alltoall operation) are strongly influenced by the congestion of network resources. In this paper we address the problem of modeling the performance of Total Exchange communication operations in grid environments. Because traditional performance models are unable to predict the real completion time of an All-to-All operation, we try to cope with this problem by identifying the factors that can interfere in both local and distant transmissions. We observe that the traditional MPI Alltoall implementation is not suited for grid environments, as it is both inefficient and hard to model. We focus therefore in an alternative algorithm for the total exchange redistribution problem. In our approach we perform communications in two different phases, aiming to minimize the number of communication steps through the wide-area network. This reduction has a direct impact on the performance modeling of the MPI Alltoall operation, as we minimize the factors that interfere with wide-area communications. Hence, we are able to define an accurate performance modeling of a total exchange between two clusters.

National audience

version étendue de l'article publié à CLUSTER2006; One of the most important collective communication patterns used in scientific applications is the complete exchange, also called All-to-All. Although efficient complete exchange algorithms have been studied for specific networks, general solutions like those available in well-known MPI distributions (e.g. the MPI_Alltoall operation) are strongly influenced by the congestion of network resources. In this paper we present an integrated approach to model the performance of the All-to-All collective operation. Our approach consists in identifying a contention signature that characterizes a given network environment, using it to augment a contention-free communication model. This approach allows an accurate prediction of the performance of the All-to-All operation over different network architectures with a small overhead. This approach is assessed by experimental results using three different network architectures, namely Fast Ethernet, Gigabit Ethernet and Myrinet.

International audience; Nowadays, the development of pervasive technologies has allowed the improvement of services availability. These services, offered by information systems (IS), are becoming more pervasive, i.e., accessed anytime, anywhere. However, those pervasive information systems (PIS) remain too complex for the user, who just wants a service satisfying his needs. This complexity requires considerable efforts from the user in order to select the most appropriate service. Thus, an important challenge in PIS is to reduce user's understanding effort. In this chapter, we propose to enhance PIS transparency and productivity through a user-centred vision based on an intentional approach. We propose an intention prediction approach. This approach allows anticipating user's future requirements, offering the most suitable service in a transparent and discrete way. This intention prediction approach is guided by the user's context. It is based on the analysis of the user's previous situations in order to learn user's behaviour in a dynamic environment.

International audience; The improvement of computational systems has been based on Moore's law and Dennard's scale, but for more than a decade it has started to fall to a standstill. To maintain the improvement new technologies are proposed, this has established a Post-Moore era. Currently there are different methodologies to evaluate emerging devices and technologies, but the results of these methodologies end up being particular solutions. This paper presents a methodology that can evaluate Post-Moore devices or technologies in an agile way, characterizing an application, choosing a device that meets its needs, selecting tests and parameters that are executed to meet the objectives set by the methodology. The results show how to evaluate the energy efficiency of these new technologies, but the scope of the methodology can cover other needs.

International audience; The quality of the harvest depends significantly on the quality of the grapes. Therefore, winemakers need to make the right decisions to obtain highquality grapes. One of the first problems is estimating the yield of the crops. It allows winemakers to respect the specific norms of their appellation (yield quota, alcohol levels, etc.). It is also necessary to organise the logistics of the harvest (start date, human resources required, transportations, etc.). Traditionally, the yield estimation is performed by collecting grapes and berries over small, randomised samples, a destructive and laborious task. This work explores how automatic data acquisition combined with artificial intelligence can drive an automated and non-destructive yield estimation, adapted to the characteristics of each vine parcel.

Paper written in Portuguese; Lançada quase simultaneamente à popularização da Internet, a linguagem Java destacou-se como uma linguagem orientada a objetos com importantes características de segurança, integração com os serviços de rede e robustez das aplicações, em ambientes multiplataforma, nas mais distribuídas e heterogêneas redes. Mais que isso, Java implementou conceitos de programação que sempre eram citados como "seria bom que existisse, mas ainda não existe nas linguagens comumente utilizadas". Cabe ressaltar, no entanto, que para a grande maioria das pessoas que já ouviram falar em Java existe uma certa "névoa de mistério", onde praticamente todos sabem que dá pra fazer animações para páginas de WWW, mas não conseguem imaginar outras aplicações ou ferramentas da linguagem. Quando foi proposta a implementação de uma situação de corrida crítica que nem o Produtor-Consumidor, nossa motivação foi a de desvendar os mistérios e capacidades da linguagem Java, aplicando-a da melhor forma possível para chegar ao nosso objetivo. Mais do que isso, devido à característica facilidade com que aplicações Java podem ser integradas às páginas na Internet, a disponibilização da aplicação e do código podem vir a fornecer recursos para pesquisa e desenvolvimento, tanto voltados à área de Sistemas Operacionais, quanto à própria linguagem Java.

International audience; This paper deals with the problem of broadcasting for clus- ter of clusters. The construction of partial minimum spanning trees being NP-complete, several heuristic algorithms have been already formulated. Many of these heuristics (like the heuristic of Kruskal) use the shortest path to connect the components of the tree. They are not relevant in case of forwarding or overlapping communication during a step of the algorithm. In this paper, we study a new heuristic for the minimum broadcasting tree and we evaluate it through simulations with different communication parameters and also through real experimentation over the Grid'5000 testbed.

International audience; The increase in computational capacities has helped in the exploration , production and research process, this has allowed the use of applications that were infeasible years ago. This increase brings us a new Era (known as Post-Moore Era) and a wide range of promising devices, devices such as Single Board Computers (SBC) or Personal Computers (PC) that achieve performance that a decade ago was only found on a Server. This work presents high performance computing devices with low monetary cost and low energy cost that meet the needs for the development of research in Artificial Intelligent (AI) applications, in-situ data analysis and simulations that can be implemented on a large scale, these devices are compared in different tests, presenting advantages such as its performance per watt consumed, smart form, among others.

International audience; Computing systems with a large number of processing units are increasingly common, both in the form of processors employing multiple execution cores (e.g., multi-core CPUs, GPUs, accelerators), or computing clusters with a large number of nodes. These many- core architectures bring up new capabilities, opportunities, as well as challenges. Adapting applications effectively use these architectures has involved a significant amount of effort, not only in code modifications, but also in developing the programming environment. However, several industry trends seem to indicate that large-scale homogeneous platforms, which contain a single type of compute engine with a relatively straightforward memory hierarchy, will soon be a thing of the past. The expansion of heterogeneity in multiple dimensions - compute, memory, storage, systems, software, applications, and usage models – leverages an extreme heterogeneity scenario that will affect both the processor industry but also the software industry, with strong repercussions on programming models, scientific workflows and data management systems. In this talk we examine how high-performance computing has changed over the last 10 years, what are the current projects towards exaflop computing, and look toward the future in terms of trends.

International audience

Paper written in Portuguese

International audience; Le Fog Computing étend le paradigme du Cloud Computing à la périphérie du réseau, en s'appuyant sur des services intelligemment distribués pour répondre au mieux aux besoins des applications. Souvent, ces services sont déployés sur des dispositifs "passerelle" caractéri-sés par une puissance de calcul relativement limitée, ce qui peut porter atteinte aux objectifs de qualité de service à l'origine de la démarche fog. Pour mieux comprendre les enjeux, cet article s'intéresse aux performances de l'accès aux données distribuées à partir de dispositifs de proxi-mité tels que les "systèmes sur puce". Des résultats expérimentaux comparant les performances d'accès aux données viennent démontrer l'intérêt de la prise en compte de ces performances lors du déploiement d'une architecture fog.

The exponential dissemination of proximity computing devices (smartphones, tablets, nanocomputers, etc.) raises important questions on how to transmit, store and analyze data in networks integrating those devices. New approaches like edge computing aim at delegating part of the work to devices in the “edge” of the network. In this article, the focus is on the use of pervasive grids to implement edge computing and leverage such challenges, especially the strategies to ensure data proximity and context awareness, two factors that impact the performance of big data analyses in distributed systems. This article discusses the limitations of traditional big data computing platforms and introduces the principles and challenges to implement edge computing over pervasive grids. Finally, using CloudFIT, a distributed computing platform, the authors illustrate the deployment of a real geophysical application on a pervasive network.