Big Data platforms for science automation

我的研究计划的目标是实现从数据处理到知识发布的大数据分析自动化。在接下来的几年里，我将重点关注这一巨大挑战的以下三个目标。(1)大数据平台之间的互通：目前用于科学的大数据平台运行在孤岛上，这阻碍了开放科学，通过限制竞争降低了平台的整体质量，并造成了对特定软件项目的技术依赖。我将设计一个连接平台的分散网络的构建块，以便数据、处理管道和分析可以在不同的平台上统一查找、访问和重用。(2)大数据分析在时间和空间上的再现性：科学正在经历严重的可再生性危机，如果解决不好，将阻碍各个学科利用获得的巨大数据财富：必须将大数据转化为可信的知识。我将开发方法来识别、量化和纠正可重复性问题，重点关注源自计算基础设施的挑战。(3)大数据计算的性能优化：虽然大数据计算对行业具有变革性，但由于缺乏适当的基准和性能优化研究，大数据技术在科学上仍未得到充分利用。我将为大型科学数据集设计和创建一个优化的、易于使用的大数据处理环境。这三个目标呼应了大数据臭名昭著的V字：互操作性满足数据多样性，可再生性以准确性为目标，性能提供速度并管理数据量。 总体而言，这一计划将加快和提高知识生产的质量。它将通过允许科学对象无缝跨平台移动、确保不同平台计算结果的一致性以及加速大数据分析来促进网络基础设施自动化。虽然这项研究的潜在应用范围涉及数据科学的所有科学学科，但我将专注于神经信息学，利用我在该领域的长期顶级合作，并利用在VIP和CBRAIN平台的开发和运营中获得的丰富经验。预计还将在IT(大数据和云)和制药行业向加拿大行业转让技术。

The objective of my research program is to automate Big Data analyses from data processing to knowledge publication. In the next years, I will focus on the following three objectives of this immense challenge. (1) Interoperability among Big Data platforms: Big Data platforms used in science currently operate in silos, which hinders open science, reduces the overall quality of platforms by limiting competition, and creates technological dependence on particular software projects. I will design the building blocks of a decentralized network connecting platforms so that data, processing pipelines and analyses can be uniformly found, accessed and reused in various platforms. (2) Reproducibility of Big Data analyses over time and space: science is going through a severe reproducibility crisis, which, if not properly addressed, will prevent various disciplines from leveraging the tremendous wealth of data that is acquired: Big Data has to be converted into trusted knowledge. I will develop methods to identify, quantify and correct reproducibility issues, focusing on the challenges that originate in the computing infrastructure. (3) Performance optimization of Big Data computations: while they have been transformational for industry, Big Data technologies remain underused in science, due to the lack of appropriate benchmarks and performance optimization studies. I will design and create an optimized, easy-to-use Big Data processing environment for large scientific datasets. These three objectives echo the notorious V's of Big Data: interoperability addresses data Variety, reproducibility targets Veracity, and performance provides Velocity and manages data Volume. Overall, this program will speed up and improve the quality of knowledge production. It will foster cyberinfrastructure automation, by allowing scientific objects to seamlessly move across platforms, by ensuring the consistency of results computed in different platforms and by accelerating Big Data analyses. While potential applications of this research span the whole spectrum of scientific disciplines engaged in data science, I will focus on neuroinformatics, exploiting my long-standing top-level collaborations in this field and leveraging the tremendous experience acquired with the development and operation of the VIP and CBRAIN platforms. Technology transfer to the Canadian industry is also expected, in the IT (Big Data and cloud) and pharmaceutical sectors.