RII Track-4: Advancing Machine Learning in Biological Oceanography Through Interdisciplinary Collaborations

University Of Alaska Fairbanks Campus, Fairbanks AK

Investigators

Abstract

Non-Technical Description Modern scientists in interdisciplinary fields like oceanography and genomics regularly generate complex datasets with billions or trillions of data points. Machine learning is revolutionizing the way these data are observed. As a merger between statistics and computer science, the techniques of machine learning have been used for decades, but recent improvements in high-performance computing, and research investment from the commercial sector have supercharged the utility of these methods. Environmental scientists have been relatively slow to take up these new methods, in part because the rapid pace of discovery makes it difficult to stay at the cutting edge, but as scientific datasets grow ever larger and more complex, it is becoming increasingly important to make use of machine learning as a way to explore and understand the world. In collaboration with colleagues at Woods Hole Oceanographic Institute, the PI aims to stimulate progress in oceanography and genomics by bringing cutting edge tools and techniques from the world of machine learning to Alaska. Acquiring these skills in parallel with an early career scientist, and applying them to ongoing research projects in the Arctic, will enable a substantially improved understanding of ongoing changes in the marine ecosystems of Alaska. Passing on this knowledge to students by designing and teaching a new course on Machine Learning in the Environmental Sciences will ensure that future generations of Alaskans have access to the skills they will need to succeed in science and beyond. Technical Description The broad goals of this visit are to advance machine learning in biological oceanography through interdisciplinary collaborations and knowledge transfer. The PI aims to leverage extensive expertise at Woods Hole Oceanographic Institution (WHOI) to advance research on Arctic marine microbial communities using machine learning methods. The WHOI Autonomous Robotics and Perception Laboratory (WARP) Lab specializes in adaptive sampling, marine robotics, computer vision, autonomous robotic exploration, semantic perception, bayesian nonparametrics, deep learning, surprise detection, and data summarization. Objectives of this project include learning the theory and concepts of machine learning in structured and informal settings, which include several common programming languages and toolkits. Using this base, the PI and a graduate trainee will apply the tools and techniques of machine learning to predict spatiotemporal distributions of Arctic marine microbes, their use and transformations of metals, and advance biological oceanographic sampling techniques. The PI will accomplish this effort by designing methods for adaptive biological sampling using flow-through systems and ocean profilers. Outcomes of the project will include the development of a new course on Machine Learning in the Environmental Sciences to be taught at the University of Alaska Fairbanks and online via the University of the Arctic. Career development and mentorship of the graduate student will include training in computing fluency in the Software Carpentry program and subsequent training as an instructor. As part of the knowledge exchange with WHOI, the PI will host an Oxford Nanopore minION sequencing workshop in Woods Hole to share this cutting edge technology, and lead cross-cutting discussions on the use of machine learning in advancing this technology. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.