Open Source Scalable Data Services and Data Integration of Geospatial and Georeferenced Data for Machine Learning

Abstract:

The availability of big data and rise of artificial intelligence is increasingly driving innovation in scientific research. For instance, the rise of deep learning has sparked tremendous improvements to state-of-the-art algorithms in the geospatial and remote sensing domain. The key to deep learning success is access to large volumes of training data. The data must be cataloged, annotated, correlated with other data sources, split into manageable-sized pieces, and made accessible to developers with sufficient computing infrastructure. Today, different types of geospatial and georeferenced big data are typically hosted in independent databases, making multi-faceted queries that correlate across data sources difficult and inefficient. To support rapid prototyping of machine learning algorithms in the geospatial domain, a framework is needed to coalesce data from multiple sources and allow developers to submit algorithms to be tested and scored for comparison.

Open source ResonantGeoData is a state-of-the-art data integration and data management solution to support data-driven algorithm development and evaluation in the geosciences domain. Resonant geodata is being actively developed using open source tools and open specifications such as OpenAPI and OGC, scalable data processing and database technologies, and best practices for software development. ResonantGeoData is enabling easier, more consistent, secure geospatial and georeferenced data access and faster turnaround for algorithmic development and evaluation. ResonantGeoData will help construct and host geospatial algorithm competitions, support algorithm prototyping, and evaluation, and provide an interface for analysts to experiment with top-performing algorithms. ResonantGeoData runs in a cloud environment for scalability in the production mode and supports the latest machine learning frameworks like PyTorch and Tensorflow in a containerized environment.