Machine Learning in Geospace

_

_

VGEM

Student Day	Sunday, July 20th, 2020 is Student Day
GEM Workshop	Monday - Thursday, July 21st - 23rd

Participants will need to [register for free here https://gemworkshop.org/pages/RegistrationVirtual.php], which is accessible through [the general meeting website https://gemworkshop.org/]. Our session agenda is posted below.

Machine Learning Session

When

Thursday, July 23, 3:00 - 4:30pm Eastern

Session Conveners

Name		Email
Matthew Argall		matthew.argall@unh.edu
Jacob Bortnik		jbortnik@gmail.com
Josh Rigler		erigler@usgs.gov
Jason Shuster		jason.r.shuster@nasa.gov

Preface

We have worked with the GEM Student Representatives to incorporate a machine learning tutorial into their Student Day. Abby Azari will give a "getting started with machine learning" tutorial.

Session Overview and Theme

Beyond the mechanics of getting started, to fully take advantage of the power of machine learning, one needs to know what is possible. This is our session's theme -- to provide an overview of what is possible. This theme encompasses five focus areas:

1. Recent developments in machine learning that enable innovative scientific research
2. Aspects of machine learning or data science that are under-utilized in the field
3. Systems/chains of machine learning models and/or the integration of machine learning into numerical simulations
4. Possibilities and opportunities that exist outside of a purely academic career path enabled by machine learning
5. Bridging the SHINE, GEM, and CEDAR communities to facilitate multi-discipline collaborations

The hope is that recent developments in ML will inspire new avenues of research and career opportunities.

Presentations and Panel Discussion

Speakers and panelists are encouraged to address any of the focus areas 1-4 and to include elements that enable focus area 5.

ML Exploratory Committee

This session is chaired by members of a SHINE/GEM/CEDAR Machine Learning Exploratory Committee. The committee is composed of members of the SHINE, GEM, and CEDAR communities. Its goal is to pursue topics in Heliophysics Machine Learning that can best be addressed through a cross-disciplinary initiative. Anyone interested can contact us for more information.

Agenda

Session

Table 1: Schedule for the Machine Learning session at VGEM. The audience can direct additional questions to the speakers during the panel discussion.

Duration	Topic	Speaker
5 min	Opening remarks about SHINE/GEM/CEDAR ML Exploratory Committee and the ML Session	Matthew Argall
15 min	Introduction to ML: "What is Possible?"	Jacob Bortnik (See Table 3)
7 min 3 min	Application(s): ML in GEM with relevance to SHINE/CEDAR	Enrico Camporeale
7 min 3 min	Application(s): ML in CEDAR with relevance to GEM
7 min 3 min	Application(s): ML in SHINE with relevance to GEM	Sophie Murray
40 min	Open Panel Discussion	See Table 2

Panel Discussion

Table 2: After a brief self-introduction, panelists to field questions in an open discussion format.

Field	Panelist	Early Career	Theme
SHINE	Wendy Carande	Y	-
SHINE	-	-	-
GEM	Simon Wing		Information Theory
GEM	Xiangning Chu	Y	Global Reconstruction
CEDAR	Bharat Kunduri	Y	Insight Industry to academia
CEDAR	-	-	-
Planetary	Abby Azari	Y	Student tutorial Physics Informed ML
Industry	-	-	-

Distributed Tutorial

Table 3: SHINE/GEM/CEDAR ML Exploratory Committee topics highlighting a subset of what is possible with machine learning. Each speaker provides an overview of the topic in ~1 minute and ~1 slide.

Speaker	Topic
Ryan McGranaghan	Community-wide inventory of assets and resources
Marcus Hughes	Exploration, visualization and handling of large observational and modeling data volumes
TBC	Understanding the potential enhancement of space weather prediction capabilities for operations and research
TBC	Creation of AI-ready data sets and quality standards
Jacob Bortnik	Different approaches to implementing “Physics-informed learning” and “gray box models”
Josh Rigler	Compression, acceleration and emulation of models and data processing
Matthew Argall	Providing paths to publication, validation, and open-code practices