The Rise of Computer-Aided Discovery in Geoscience

Abstract:

Next-generation Geoscience will need to handle rapidly growing data volumes and exploration of complex phenomena challenging human cognitive limits. With instruments digitizing large amounts of sensor data from many sources, the scientific discovery process becomes a large-scale search process. However, insight generation is still a key problem and is especially complex in Geoscience, particularly when exploratory studies involve fusion of large data from various instruments in a manual labor-intensive manner.

We propose an approach for a computer-aided discovery infrastructure that automatically explores the connection between physics models and empirical data to accelerate the pace of new discoveries. The approach uses (1) A system engaging scientists to programmatically express hypothesized Geoscience scenarios, constraints, and model variations, so as to automatically explore and evaluate the combinatorial search space of possible explanations in parallel on a variety of data sets. This automated system employs machine learning to support algorithmic choice and workflow reconfiguration allowing systematic pruning of the search space of applied algorithms and parameters based on historical results. (2) A cloud-based environment allowing scientists to conduct powerful exploratory analyses on large data sets that reside in data centers. Various search modes are provided, including a mode where scientists can iteratively guide the search based on intermediate results. This functionality directs the system to identify more Geospace features that are analogous or related in various ways. (3) Scientist input is used to configure programmable crawlers that automate and scale the search for interesting phenomena on cloud-based infrastructures. We discuss various application scenarios to show the impact of workflow configuration on scientific feature detection.

Acknowledgements. We acknowledge support from NSF ACI-1442997 and NASA AIST NNX15AG84G (PI: V. Pankratius).