[Esip-machinelearning] ESIP Machine Learning Cluster Newsletter - March 2020

[Yuhan Douglas Rao]

Dear all,

Hope you are all doing well during this special time. Here is our newest
edition of ESIP Machine Learning Cluster Newsletter (pdf version also
attached). Hope this can help our community to stay connected despite rapid
changing conditions! As usual, if you have any questions and suggestions,
please feel free to reach out to Anne and me.

Stay well!
Kind regards,
Yuhan (Douglas) Rao
ESIP Community Fellow for Machine Learning Cluster

Earth Science Information Partners

Machine Learning Cluster Newsletter

March 17, 2020

------------------------------

Message from the Chair

Thank you for taking a look at our Machine Learning (ML) Cluster
newsletter! There’s a lot of ML activity in and around ESIP, listed below.
Aspects of machine learning represented here include:

   -

   the application of ML in various science domains;
   -

   educational opportunities about a variety of ML tools;
   -

   ML funding opportunities;
   -

   understanding the conclusions of a trained ML model;
   -

   agency and other organizational efforts to integrate ML: practices,
   strategies;

In February 2019, the American Association for the Advancement of Science
(AAAS) published an article titled, “Can we trust scientific discoveries
made using machine learning?
<https://eurekalert.org/pub_releases/2019-02/ru-cwt021119.php>” The answer
was, “Not without checking.” Why is that?

Rather than following explicit instructions, ML algorithms learn from data.
These algorithms must always return a predication - the result of “I don’t
know,” or “none of the above” is not an option. This can lead to
uncertainty and results that may not be corroborated. Often, ML algorithms
are obscure even to their creators. While this may be acceptable in some
domains, such as product recommendation, the ESIP community well
understands the need for transparency, understanding, and reproducibility
in science.

Cultural methods to manage this concern include the development of ML
standards, best practices, Technical Readiness Levels, etc., as NOAA is
doing (below). NIST
<https://www.nist.gov/news-events/news/2019/07/nist-releases-draft-plan-federal-engagement-ai-standards-development>
and even the Catholic Church
<https://www.theverge.com/2020/2/28/21157667/catholic-church-ai-regulations-protect-people-ibm-microsoft-sign>are
also proposing AI standards. (I have yet to find similar efforts for NASA
or USGS. Anyone?)

Technical solutions can involve human-mediated construction of
understandable ML models or post hoc model analysis. Here, I would
particularly like to call out the upcoming 2020 Geosemantics Symposium
(below), occurring in coordination with the ESIP Summer Meeting
<https://2020esipsummermeeting.sched.com/info>. This one day event brings
our cluster together with the Semantic Technologies Committee. This summer
we focus on Explainable AI (xAI)
<https://en.wikipedia.org/wiki/Explainable_artificial_intelligence>, which
aims to address how decisions are made by, in our case, ML systems. Stay
tuned for more details regarding date, time, and agenda.

Notable award. At the 2020 Winter Meeting in January, our Cluster Community
Fellow, Yuhan Rao, won the ESIP Catalyst Award
<https://www.esipfed.org/press-releases/peer-recognition-2020-esip-winter-meeting>.
That award “honors those who have brought about positive change in ESIP and
inspired others to take action in the past year”. I can personally attest
that Yuhan, the author and instigator of this newsletter, has brought tons
of positive energy (to mix physical phenomenon) to our cluster and well
beyond. He has regularly inspired me to get up and do something (hopefully
for the better.) Congratulations, Yuhan! And, thank you!

Anne Wilson, Ronin Institute

March 9, 2020
Inside the cluster

Highlights of ESIP-ML and cluster members’ activities

   -

   2020 ESIP Winter meeting. Last month’s ESIP Winter Meeting has again
   highlighted the rapid growing interests of machine learning and artificial
   intelligence from the Earth sciences information community. There are
   multiple sessions featuring different aspects of ML/AI. For example, “AI
   for augmenting geospatial information discovery” (watch recording here
   <https://www.youtube.com/watch?v=W0q8WiMw9Hs&feature=youtu.be>) featured
   different use cases of AI technologies for agriculture, climate, and social
   good etc; Agriculture & Climate cluster
   <http://wiki.esipfed.org/index.php/Agriculture_and_Climate> initiated a
   new effort on “automated agriculture using AI” (watch recording here
   <https://youtu.be/GhnSINRFNBg>). Our cluster also hosted a two-session
   workshop focusing on the topic of training data. This workshop introduced
   three existing tools/platforms for generating training data -- Image
   Labeler <https://labeler.nasa-impact.net/>, LabelImg
   <https://github.com/tzutalin/labelImg>, Bokeh <https://bokeh.org/>, and
   hands-on activities to use LabelImg and Bokeh to generate labelled
data (watch
   recording here <https://youtu.be/3ufBOoD3M1E>). These three tools
   provide great potential to generate different types of training data via
   well developed open source ecosystems. Although the winter meeting was just
   concluded, we have already been thinking about the topics for the summer
   meeting in Burlington, VT. We would love to hear your ideas and suggestions
   and hope to make them happen for the summer meeting.
   -

   2020 GeoSemantics Symposium/workshop. Together with the Semantic
   Technologies Committee
   <http://wiki.esipfed.org/index.php/Semantic_Technologies>, we are
   proposing to jointly organize the summer GeoSemantic Symposium to coincide
   with the ESIP summer meeting. The proposed theme for this year’s symposium
   is leveraging explainable AI (xAI) to support decision making with Earth
   science data. The proposed theme, echoing ESIP’s strategic goal of putting
   data to work, aims to bring diverse stakeholders (technologists, data
   providers, and decision makers) together and use semantic technologies to
   increase our confidence in ML-supported decision making processes. At this
   early planning stage, we are looking for use cases from the ESIP
   community that fits the theme and develop an engaging program to put
   data to work!
   -

   AI for Agriculture (Agro AI). The Agriculture and Climate cluster
   <http://wiki.esipfed.org/index.php/Agriculture_and_Climate> (ACC) just
   started the discussion on AI for Agriculture (Agro AI) effort . We believe
   this can be a great opportunity to foster stronger connections and
   collaboration across ESIP. If you have ideas or successful examples in this
   area, we encourage you to share it with members from both clusters and
   participate in the ACC’s discussion on this new effort. You can
contact William
   (Bill) Teng (william.l.teng at nasa.gov) and Brian Wee (
   bwee at massiveconnections.com), ACC’s co-chairs, to learn more about this
   exciting new effort!

All about Machine Learning

Information relevant to ESIP-ML community

   -

   NOAA’s AI Strategy. NOAA Research Council (NRC)
   <https://nrc.noaa.gov/NOAA-Science-Technology-Focus-Areas> has recently
   released NOAA’s new strategies in four key science and technology focus
   areas. Among these key focus areas, NOAA’s AI strategy
   <https://nrc.noaa.gov/LinkClick.aspx?fileticket=0I2p2-Gu3rA%3d&tabid=91&portalid=0>
   aims to “dramatically expand the application of artificial intelligence
   (AI) in every NOAA mission area by improving the efficiency, effectiveness,
   and coordination of AI development and usage across the agency”. NOAA’s new
   strategy together with other federal agencies' efforts can boost Earth
   sciences community’s effort to embrace ML/AI as tools to advance scientific
   discovery and support societal relevant decision making process.
   -

   Radiant Earth/NASA ML workshop. During January 21-23, 2020, Radiant
   Earth Foundation hosted an international workshop
   <https://www.radiant.earth/events/nasa-ml-2020/> on leveraging ML and
   NASA Earth observations (EO) to address environmental challenges (ML4EO).
   The recordings of the workshop have been archived on Radiant Earth
   Foundation’s YouTube channel
   <https://www.youtube.com/playlist?list=PL3QzFgBMGnbQRa8uHP0_C_P2Fl5GIBxmn>
   .
   -

   Microsoft AI for Earth Grant. Microsoft’s AI for Earth
   <https://www.microsoft.com/en-us/ai/ai-for-earth-grants> program
   provides cloud computing resources and AI tools for researchers and
   stakeholders (NGOs, local governments, etc) to solve environmental
   challenges. The grant provides Azure compute credits for selected projects
   to solve four categories of environmental challenges -- climate,
   agriculture, biodiversity and water. The grant has four application
   deadlines each year and the upcoming application deadline is April 6,
   2020.
   -

   National Geographic AI for Earth Innovation Grants. To address the many
   pressing scientific questions and challenges facing our planet, we must
   increase global understanding of how human activity is affecting natural
   systems and create a community of change, driven by data and cutting-edge
   technology. Modern technologies, such as satellite imaging, bioacoustic
   monitoring, environmental DNA, and genomics, can capture data at a global
   scale, but also produce massive, complex data sets. Artificial intelligence
   (AI) and cloud computing can capitalize on the potential of such data,
   leading to faster and more meaningful insights and creating the opportunity
   for transformative solutions. Learn more here
   <https://www.nationalgeographic.org/funding-opportunities/grants/what-we-fund/ai-earth-innovation/>.
   The deadline for the RFP is July 22, 2020.

------------------------------

You are receiving this newsletter because you are part of the ESIP Machine
Learning Cluster. If you have any news want to share or questions, please
contact Anne Wilson (anne.wilson *at* ronin.org) and Yuhan Rao (yuhan.rao
*at* gmail.com).
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.esipfed.org/pipermail/esip-machinelearning/attachments/20200317/66deae89/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ML cluster newsletter - 2020 March - final.pdf
Type: application/pdf
Size: 460687 bytes
Desc: not available
URL: <http://lists.esipfed.org/pipermail/esip-machinelearning/attachments/20200317/66deae89/attachment-0001.pdf>