Software per l'analisi di Big Data su più livelli e sua applicazione a tracce di Google

Con l'avanzare della tecnologia, i Big Data hanno assunto un ruolo importante. In questo lavoro è stato implementato, in linguaggio Java, un software volto alla analisi dei Big Data mediante R e Hadoop/MapReduce. Il software è stato utilizzato per analizzare le tracce rilasciate da Google, riguardanti il funzionamento dei suoi data center.

A BAYESIAN HIERARCHICAL FRAMEWORK FOR SPATIAL MODELING OF fMRI DATA

Functional neuroimaging techniques enable investigations into the neural basis of human cognition, emotions, and behaviors. In practice, applications of functional magnetic resonance imaging (fMRI) have provided novel insights into the neuropathophysiology of major psychiatric,neurological, and substance abuse disorders, as well as into the neural responses to their treatments. Modern activation studies often compare localized task-induced changes in brain activity between experimental groups. One may also extend voxel-level analyses by simultaneously considering the ensemble of voxels constituting an anatomically defined region of interest (ROI) or by considering means or quantiles of the ROI. In this work we present a Bayesian extension of voxel-level analyses that offers several notable benefits. First, it combines whole-brain voxel-by-voxel modeling and ROI analyses within a unified framework. Secondly, an unstructured variance/covariance for regional mean parameters allows for the study of inter-regional functional connectivity, provided enough subjects are available to allow for accurate estimation. Finally, an exchangeable correlation structure within regions allows for the consideration of intra-regional functional connectivity. We perform estimation for our model using Markov Chain Monte Carlo (MCMC) techniques implemented via Gibbs sampling which, despite the high throughput nature of the data, can be executed quickly (less than 30 minutes). We apply our Bayesian hierarchical model to two novel fMRI data sets: one considering inhibitory control in cocaine-dependent men and the second considering verbal memory in subjects at high risk for Alzheimer’s disease. The unifying hierarchical model presented in this manuscript is shown to enhance the interpretation content of these data sets.

Spatially explicit estimates of North Atlantic albacore tuna (Thunnus alalunga) biomass in the North Atlantic for the period 1956-2010 - Gridded data product (NetCDF)

The development of the ecosystem approach and models for the management of ocean marine resources requires easy access to standard validated datasets of historical catch data for the main exploited species, together with the model estimates achieved from these data, allowing models inter-comparison and evaluation of model skills. North Atlantic albacore tuna is exploited all year round by longline and in summer and autumn by surface fisheries and fishery statistics compiled by the International Commission for the Conservation of Atlantic Tunas (ICCAT). Catch and effort with geographical coordinates at monthly spatial resolution of 1° or 5° squares were extracted for this species with a careful definition of fisheries and data screening. Length frequencies of catch were also extracted according to the definition of fisheries for the period 1956-2010. Using these data, an application of the spatial ecosystem and population dynamics model (SEAPODYM) was developed for the North Atlantic albacore population and fisheries and provided the first spatially explicit estimate of albacore density in the North Atlantic by life stage. These densities by life stage (larval recruits, young immature fish adult mature fish and total biomass) are provided in gridded file (Netcdf) at resolution of 2° x 2° x month.

Simulating Data Journalism to Communicate Hydrological Information from Sensor Networks

Presenting relevant information via web-based user friendly interfac- es makes the information more accessible to the general public. This is especial- ly useful for sensor networks that monitor natural environments. Adequately communicating this type of information helps increase awareness about the limited availability of natural resources and promotes their better use with sus- tainable practices. In this paper, I suggest an approach to communicating this information to wide audiences based on simulating data journalism using artifi- cial intelligence techniques. I analyze this approach by describing a pioneer knowledge-based system called VSAIH, which looks for news in hydrological data from a national sensor network in Spain and creates news stories that gen- eral users can understand. VSAIH integrates artificial intelligence techniques, including a model-based data analyzer and a presentation planner. In the paper, I also describe characteristics of the hydrological national sensor network and the technical solutions applied by VSAIH to simulate data journalism.

Meter data management for smart monitoring power networks

The electrical power distribution and commercialization scenario is evolving worldwide, and electricity companies, faced with the challenge of new information requirements, are demanding IT solutions to deal with the smart monitoring of power networks. Two main challenges arise from data management and smart monitoring of power networks: real-time data acquisition and big data processing over short time periods. We present a solution in the form of a system architecture that conveys real time issues and has the capacity for big data management.

GRS Results for the Burnup Pin-cell Benchmark Propagation of Cross-Section, Fission Yields and Decay Data Uncertainties

GRS Results for the Burnup Pin-cell Benchmark Propagation of Cross-Section, Fission Yields and Decay Data Uncertainties

Lagrangian tools and the assessment of the predictive capacity of geophysical data sets

We examine, with recently developed Lagrangian tools, altimeter data and numerical simulations obtained from the HYCOM model in the Gulf of Mexico. Our data correspond to the months just after the Deepwater Horizon oil spill in the year 2010. Our Lagrangian analysis provides a skeleton that allows the interpretation of transport routes over the ocean surface. The transport routes are further verified by the simultaneous study of the evolution of several drifters launched during those months in the Gulf of Mexico. We find that there exist Lagrangian structures that justify the dynamics of the drifters, although the agreement depends on the quality of the data. We discuss the impact of the Lagrangian tools on the assessment of the predictive capacity of these data sets.

The Imaging and Slitless Spectroscopy Instrument for Surveys (ISSIS): expected radiometric performance, operation modes and data handling

ISSIS is the instrument for imaging and slitless spectroscopy on-board WSO-UV. In this article, a detailed comparison between ISSIS expected radiometric performance and other ultraviolet instruments is shown. In addition, we present preliminary information on the performance verification tests and on the foreseen procedures for in-flight operation and data handling.

Formal Verification of Temporal Properties for Parametrized Concurrent Programs and Concurrent Data Structures

Los tipos de datos concurrentes son implementaciones concurrentes de las abstracciones de datos clásicas, con la diferencia de que han sido específicamente diseñados para aprovechar el gran paralelismo disponible en las modernas arquitecturas multiprocesador y multinúcleo. La correcta manipulación de los tipos de datos concurrentes resulta esencial para demostrar la completa corrección de los sistemas de software que los utilizan. Una de las mayores dificultades a la hora de diseñar y verificar tipos de datos concurrentes surge de la necesidad de tener que razonar acerca de un número arbitrario de procesos que invocan estos tipos de datos de manera concurrente. Esto requiere considerar sistemas parametrizados. En este trabajo estudiamos la verificación formal de propiedades temporales de sistemas concurrentes parametrizados, poniendo especial énfasis en programas que manipulan estructuras de datos concurrentes. La principal dificultad a la hora de razonar acerca de sistemas concurrentes parametrizados proviene de la interacción entre el gran nivel de concurrencia que éstos poseen y la necesidad de razonar al mismo tiempo acerca de la memoria dinámica. La verificación de sistemas parametrizados resulta en sí un problema desafiante debido a que requiere razonar acerca de estructuras de datos complejas que son accedidas y modificadas por un numero ilimitado de procesos que manipulan de manera simultánea el contenido de la memoria dinámica empleando métodos de sincronización poco estructurados. En este trabajo, presentamos un marco formal basado en métodos deductivos capaz de ocuparse de la verificación de propiedades de safety y liveness de sistemas concurrentes parametrizados que manejan estructuras de datos complejas. Nuestro marco formal incluye reglas de prueba y técnicas especialmente adaptadas para sistemas parametrizados, las cuales trabajan en colaboración con procedimientos de decisión especialmente diseñados para analizar complejas estructuras de datos concurrentes. Un aspecto novedoso de nuestro marco formal es que efectúa una clara diferenciación entre el análisis del flujo de control del programa y el análisis de los datos que se manejan. El flujo de control del programa se analiza utilizando reglas de prueba y técnicas de verificación deductivas especialmente diseñadas para lidiar con sistemas parametrizados. Comenzando a partir de un programa concurrente y la especificación de una propiedad temporal, nuestras técnicas deductivas son capaces de generar un conjunto finito de condiciones de verificación cuya validez implican la satisfacción de dicha especificación temporal por parte de cualquier sistema, sin importar el número de procesos que formen parte del sistema. Las condiciones de verificación generadas se corresponden con los datos manipulados. Estudiamos el diseño de procedimientos de decisión especializados capaces de lidiar con estas condiciones de verificación de manera completamente automática. Investigamos teorías decidibles capaces de describir propiedades de tipos de datos complejos que manipulan punteros, tales como implementaciones imperativas de pilas, colas, listas y skiplists. Para cada una de estas teorías presentamos un procedimiento de decisión y una implementación práctica construida sobre SMT solvers. Estos procedimientos de decisión son finalmente utilizados para verificar de manera automática las condiciones de verificación generadas por nuestras técnicas de verificación parametrizada. Para concluir, demostramos como utilizando nuestro marco formal es posible probar no solo propiedades de safety sino además de liveness en algunas versiones de protocolos de exclusión mutua y programas que manipulan estructuras de datos concurrentes. El enfoque que presentamos en este trabajo resulta ser muy general y puede ser aplicado para verificar un amplio rango de tipos de datos concurrentes similares. Abstract Concurrent data types are concurrent implementations of classical data abstractions, specifically designed to exploit the great deal of parallelism available in modern multiprocessor and multi-core architectures. The correct manipulation of concurrent data types is essential for the overall correctness of the software system built using them. A major difficulty in designing and verifying concurrent data types arises by the need to reason about any number of threads invoking the data type simultaneously, which requires considering parametrized systems. In this work we study the formal verification of temporal properties of parametrized concurrent systems, with a special focus on programs that manipulate concurrent data structures. The main difficulty to reason about concurrent parametrized systems comes from the combination of their inherently high concurrency and the manipulation of dynamic memory. This parametrized verification problem is very challenging, because it requires to reason about complex concurrent data structures being accessed and modified by threads which simultaneously manipulate the heap using unstructured synchronization methods. In this work, we present a formal framework based on deductive methods which is capable of dealing with the verification of safety and liveness properties of concurrent parametrized systems that manipulate complex data structures. Our framework includes special proof rules and techniques adapted for parametrized systems which work in collaboration with specialized decision procedures for complex data structures. A novel aspect of our framework is that it cleanly differentiates the analysis of the program control flow from the analysis of the data being manipulated. The program control flow is analyzed using deductive proof rules and verification techniques specifically designed for coping with parametrized systems. Starting from a concurrent program and a temporal specification, our techniques generate a finite collection of verification conditions whose validity entails the satisfaction of the temporal specification by any client system, in spite of the number of threads. The verification conditions correspond to the data manipulation. We study the design of specialized decision procedures to deal with these verification conditions fully automatically. We investigate decidable theories capable of describing rich properties of complex pointer based data types such as stacks, queues, lists and skiplists. For each of these theories we present a decision procedure, and its practical implementation on top of existing SMT solvers. These decision procedures are ultimately used for automatically verifying the verification conditions generated by our specialized parametrized verification techniques. Finally, we show how using our framework it is possible to prove not only safety but also liveness properties of concurrent versions of some mutual exclusion protocols and programs that manipulate concurrent data structures. The approach we present in this work is very general, and can be applied to verify a wide range of similar concurrent data types.

Semi-supervised learning of hierarchical latent trait models for data visualisation

An interactive hierarchical Generative Topographic Mapping (HGTM) ¸iteH_GTM has been developed to visualise complex data sets. In this paper, we build a more general visualisation system by extending the HGTM visualisation system in 3 directions: bf (1) We generalize HGTM to noise models from the exponential family of distributions. The basic building block is the Latent Trait Model (LTM) developed in ¸iteKaban_pami. bf (2) We give the user a choice of initializing the child plots of the current plot in either em interactive, or em automatic mode. In the interactive mode the user interactively selects ``regions of interest'' as in ¸iteH_GTM, whereas in the automatic mode an unsupervised minimum message length (MML)-driven construction of a mixture of LTMs is employed. bf (3) We derive general formulas for magnification factors in latent trait models. Magnification factors are a useful tool to improve our understanding of the visualisation plots, since they can highlight the boundaries between data clusters. The unsupervised construction is particularly useful when high-level plots are covered with dense clusters of highly overlapping data projections, making it difficult to use the interactive mode. Such a situation often arises when visualizing large data sets. We illustrate our approach on a toy example and apply our system to three more complex real data sets.

Maintaining legacy data: saving Belfast Harbour (UK) tide-gauge data (1902-2010)

Tide gauge data are identified as legacy data given the radical transition between observation method and required output format associated with tide gauges over the 20th-century. Observed water level variation through tide-gauge records is regarded as the only significant basis for determining recent historical variation (decade to century) in mean sea-level and storm surge. There are limited tide gauge records that cover the 20th century, such that the Belfast (UK) Harbour tide gauge would be a strategic long-term (110 years) record, if the full paper-based records (marigrams) were digitally restructured to allow for consistent data analysis. This paper presents the methodology of extracting a consistent time series of observed water levels from the 5 different Belfast Harbour tide gauges’ positions/machine types, starting late 1901. Tide-gauge data was digitally retrieved from the original analogue (daily) records by scanning the marigrams and then extracting the sequential tidal elevations with graph-line seeking software (Ungraph™). This automation of signal extraction allowed the full Belfast series to be retrieved quickly, relative to any manual x–y digitisation of the signal. Restructuring variably lengthed tidal data sets to a consistent daily, monthly and annual file format was undertaken by project-developed software: Merge&Convert and MergeHYD allow consistent water level sampling both at 60 min (past standard) and 10 min intervals, the latter enhancing surge measurement. Belfast tide-gauge data have been rectified, validated and quality controlled (IOC 2006 standards). The result is a consistent annual-based legacy data series for Belfast Harbour that includes over 2 million tidal-level data observations.

Processing Argo oxygen data at the DAC level cookbook

This document does NOT address the issue of oxygen data quality control (either real-time or delayed mode). As a preliminary step towards that goal, this document seeks to ensure that all countries deploying floats equipped with oxygen sensors document the data and metadata related to these floats properly. We produced this document in response to action item 14 from the AST-10 meeting in Hangzhou (March 22-23, 2009). Action item 14: Denis Gilbert to work with Taiyo Kobayashi and Virginie Thierry to ensure DACs are processing oxygen data according to recommendations. If the recommendations contained herein are followed, we will end up with a more uniform set of oxygen data within the Argo data system, allowing users to begin analysing not only their own oxygen data, but also those of others, in the true spirit of Argo data sharing. Indications provided in this document are valid as of the date of writing this document. It is very likely that changes in sensors, calibrations and conversions equations will occur in the future. Please contact V. Thierry (vthierry@ifremer.fr) for any inconsistencies or missing information. A dedicated webpage on the Argo Data Management website (www) contains all information regarding Argo oxygen data management : current and previous version of this cookbook, oxygen sensor manuals, calibration sheet examples, examples of matlab code to process oxygen data, test data, etc..

A novel datacenter-oriented data placement strategy of scientific workflow in hybrid cloud

Scientific workflow is a complicated data intensive application. How to achieve an effective data placement schema in hybrid cloud environment has become a crucial issue nowadays, especially with the new challenges brought by the security issues. Traditional data placement strategies usually adopt load balancing-based partition model to allocate datasets. Although these data placement schemas can have good performance in load balancing, their data transfer time may not be optimal. In contrast to traditional strategies, this paper focuses on the hybrid cloud environment and proposes a data dependency destruction-based partition model to achieve the minimal data dependency destruction partition. In addition, it presents a novel datacenter-oriented data placement strategy. This strategy allocates high dependency datasets to one datacenter according to the new partition model and thus significantly reduces data transfer time between datacenters. Experimental results show that the proposed strategy can effectively reduce data transfer time during workflow's execution.

Improving Resource Management in Virtualized Data Centers using Application Performance Models

The rapid growth of virtualized data centers and cloud hosting services is making the management of physical resources such as CPU, memory, and I/O bandwidth in data center servers increasingly important. Server management now involves dealing with multiple dissimilar applications with varying Service-Level-Agreements (SLAs) and multiple resource dimensions. The multiplicity and diversity of resources and applications are rendering administrative tasks more complex and challenging. This thesis aimed to develop a framework and techniques that would help substantially reduce data center management complexity. We specifically addressed two crucial data center operations. First, we precisely estimated capacity requirements of client virtual machines (VMs) while renting server space in cloud environment. Second, we proposed a systematic process to efficiently allocate physical resources to hosted VMs in a data center. To realize these dual objectives, accurately capturing the effects of resource allocations on application performance is vital. The benefits of accurate application performance modeling are multifold. Cloud users can size their VMs appropriately and pay only for the resources that they need; service providers can also offer a new charging model based on the VMs performance instead of their configured sizes. As a result, clients will pay exactly for the performance they are actually experiencing; on the other hand, administrators will be able to maximize their total revenue by utilizing application performance models and SLAs. This thesis made the following contributions. First, we identified resource control parameters crucial for distributing physical resources and characterizing contention for virtualized applications in a shared hosting environment. Second, we explored several modeling techniques and confirmed the suitability of two machine learning tools, Artificial Neural Network and Support Vector Machine, to accurately model the performance of virtualized applications. Moreover, we suggested and evaluated modeling optimizations necessary to improve prediction accuracy when using these modeling tools. Third, we presented an approach to optimal VM sizing by employing the performance models we created. Finally, we proposed a revenue-driven resource allocation algorithm which maximizes the SLA-generated revenue for a data center.

Data science

Guest lecture COMP1205, fundamentals and applications of data science