Architecture, techniques and models for enabling Data Science in the Gaia Mission Archive
| Autor: | Tapiador de Pedro, Daniel |
|---|---|
| Přispěvatelé: | Huedo Cuesta, Eduardo, Sarro Baro, Luis Manuel |
| Jazyk: | angličtina |
| Rok vydání: | 2017 |
| Předmět: | |
| Zdroj: | E-Prints Complutense. Archivo Institucional de la UCM instname E-Prints Complutense: Archivo Institucional de la UCM Universidad Complutense de Madrid |
| Popis: | The massive amounts of data that the world produces every day pose new challenges to modern societies in terms of how to leverage their inherent value. Social networks, instant messaging, video, smart devices and scientific missions are just mere examples of the vast number of sources generating data every second. As the world becomes more and more digitalized, new needs arise for organizing, archiving, sharing, analyzing, visualizing and protecting the ever-increasing data sets, so that we can truly develop into a data-driven economy that reduces inefficiencies and increases sustainability, creating new business opportunities on the way. Traditional approaches for harnessing data are not suitable any more as they lack the means for scaling to the larger volumes in a timely and cost efficient manner. This has somehow changed with the advent of Internet companies like Google and Facebook, which have devised new ways of tackling this issue. However, the variety and complexity of the value chains in the private sector as well as the increasing demands and constraints in which the public one operates, needs an ongoing research that can yield newer strategies for dealing with data, facilitate the integration of providers and consumers of information, and guarantee a smooth and prompt transition when adopting these cutting-edge technological advances. This thesis aims at providing novel architectures and techniques that will help perform this transition towards Big Data in massive scientific archives. It highlights the common pitfalls that must be faced when embracing it and how to overcome them, especially when the data sets, their transformation pipelines and the tools used for the analysis are already present in the organizations. Furthermore, a new perspective for facilitating a smoother transition is laid out. It involves the usage of higher-level and use case specific frameworks and models, which will naturally bridge the gap between the technological and scientific domains. This alternative will effectively widen the possibilities of scientific archives and therefore will contribute to the reduction of the time to science. The research will be applied to the European Space Agency cornerstone mission Gaia, whose final data archive will represent a tremendous discovery potential. It will create the largest and most precise three dimensional chart of our galaxy (the Milky Way), providing unprecedented position, parallax and proper motion measurements for about one billion stars. The successful exploitation of this data archive will depend to a large degree on the ability to offer the proper architecture, i.e. infrastructure and middleware, upon which scientists will be able to do exploration and modeling with this huge data set. In consequence, the approach taken needs to enable data fusion with other scientific archives, as this will produce the synergies leading to an increment in scientific outcome, both in volume and in quality. The set of novel techniques and frameworks presented in this work addresses these issues by contextualizing them with the data products that will be generated in the Gaia mission. All these considerations have led to the foundations of the architecture that will be leveraged by the Science Enabling Applications Work Package. Last but not least, the effectiveness of the proposed solution will be demonstrated through the implementation of some ambitious statistical problems that will require significant computational capabilities, and which will use Gaia-like simulated data (the first Gaia data release has recently taken place on September 14th, 2016). These ambitious problems will be referred to as the Grand Challenge, a somewhat grandiloquent name that consists in inferring a set of parameters from a probabilistic point of view for the Initial Mass Function (IMF) and Star Formation Rate (SFR) of a given set of stars (with a huge sample size), from noisy estimates of their masses and ages respectively. This will be achieved by using Hierarchical Bayesian Modeling (HBM). In principle, the HBM can incorporate stellar evolution models to infer the IMF and SFR directly, but in this first step presented in this thesis, we will start with a somewhat less ambitious goal: inferring the PDMF and PDAD. Moreover, the performance and scalability analyses carried out will also prove the suitability of the models for the large amounts of data that will be available in the Gaia data archive. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|