A guest post by Jessica Luo, Ph.D, Marine Biology and Fisheries at the Rosenstiel School of Marine and Atmospheric Science, University of Miami.
The most recent Paris conference on climate change, COP21, underscored how critical it is to understand and manage the impacts of climate change in all aspects of the human and earth ecosystem.
A key part of that is the oceans. Understanding how the base of the food web responds to, and also potentially modifies, the environment, is a critical indicator. The first Data Science Bowl gave us a powerful tool to do just that. The Bowl yielded a fundamentally different kind of image-classification technique to the field of plankton imaging.
Why plankton? You might remember that plankton ecology answers wide-ranging questions, from fisheries to how the ocean responds to climate change.
Within marine ecology, previous methods for analyzing plankton imaging data ranged from time-consuming manual analyses to semi-automated classification, which used classifiers such as Random Forest and Support Vector Machine, but still required manual verification of the images. The scientific process was often delayed by months to years as a result.
With any one of the top solutions from the Data Science Bowl, all using Convolutional Neural Nets, we would have gotten significant increases in classification accuracy, but the Data Science Bowl specifically allowed us to:
- Quantify and comprehensively evaluate a suite of classification schemes to find the very best ones
- Connect our scientific team with the top machine learning and computer vision researchers in the world; this has cascaded into a suite of activity at Oregon State University, and also in France at the University of Pierre and Marie Curie, in quickly utilizing the solutions from the competition in a variety of plankton imaging systems
Currently, we are working with one of the authors of a top-placing team, Dr. Benjamin Graham of the University of Warwick, on implementation of his technique, “Sparse Convolutional Neural Nets.”
The timing for this breakthrough was perfect—within the plankton imaging/ecology field, our collective needs had progressed to a point where we were no longer able to solve them in-house, and needed to quickly find outside solutions. The Data Science Bowl connected two rapidly changing fields (plankton ecology and computer vision) to solve a global problem.
Last year’s Data Science Bowl played a significant part in facilitating rapid analysis of the marine environment—which is essential for advancing knowledge and providing a better earth for generations to come. I can’t wait to see what the impact of this year’s competition can be for the field of medicine!
—Written by Jessica Luo