Hackweek 2020 Teams Use Machine
Learning and GPUs to Analyze Heliophysics Data


Above: The Official Heliophysics Hackweek 2020 (HW20) GitHub site.



Above: Original Members and supporters of the HW20 "Coronal Holes" Team. Clockwise from top left: Silvina Guidoni, Ajay Kumar Tiwari, Matthew Penn, Emily Mason, Luisa Capannolo, Yari Collado-Vega, Michael Kirk, Tanmoy Samanta, James Staeben, Benoit Tremblay, and Andong Hu.

A virtual hackweek focused on heliophysics took place from August 20–28, 2020, hosted by the NASA Center for Climate Simulation (NCCS) at NASA Goddard and co-sponsored by NVIDIA, with strong support from the University of Washington eScience Institute. NASA and NCCS would also like to acknowledge the support of the University of Maryland Geographical Sciences Department, who offered sponsorship and a physical site for the event that we were unable to use due to the COVID-19 pandemic.

Ultimately, the main goal of Heliophysics Hackweek 2020 (HW20) was achieved: the collaborative hackweek was held on five days spread across a two-week period, preceded by a virtual HW20 Project Pitchfest on July 20. The participants included dozens of heliophysics scientists and data scientists from across NASA Goddard Space Flight Center, researchers from over a dozen universities, and several participants from agencies including the National Oceanic and Atmospheric Administration (NOAA), the US Naval Research Laboratory (NRL), and the Universities Space Research Association (USRA). Five project teams collaborated with one another before and during the two weeks of the event on five down-selected, real-world, intensive artificial intelligence/machine learning (AI/ML) projects using NVIDIA GPUs and NCCS resources such as containers and Jupyter Notebooks.

All applicants received practical ML training and experience using GPUs through training modules delivered by NCCS and NVIDIA staff during HW20 Week One and through hands-on ML in teams during Week Two using pre-optimized solar datasets, Jupyter Notebooks, and NVIDIA-provided GPUs. During the event preparation stage and the Pitchfest, training days, group meetings, and the final results presentation, a number of potential new NCCS users worked closely with NCCS and science-managed cloud environment (SMCE) staff, learning about Discover, ADAPT, and other NCCS resources, as planned. Beyond meeting the stated goals and objectives, several hackweek teams plan to continue working together and publish hackweek coding results.


Above: Members and supporters of the HW20 "Atmospheric Relativistic Electron Precipitation Predictions" Team. Clockwise from top left: Mike Shumko, Sapna Shekhar, Luisa Capannolo, Aziz Diaby, Steven Forsyth, Jian Li, David Linko, and Barbara Thompson.



Above: Members and supporters of the HW20 "Fluid Simulations" Team. Clockwise from top left: Yihua Zheng, Lutz Rastaetter, Sheng Huang, Sean Keefe, Ryan McGranaghan, Jian Li, Liang Wang, and David Hall.

Above: NASA HW20 sponsor and planner Mark Carroll and NVIDIA team. Clockwise from top left: Steven Forsyth, Jason Tichy, Matthew Nicely, Mark Carroll, Jim Hooks, David Hall, Zahra Ronaghi, Mark Carroll, and Matthew Penn.

As technology evolves, it creates opportunities and generates challenges. Scientists need to be at the forefront of technology to remain competitive and ensure they are doing the best science possible. Staying abreast of technologies can be accomplished by attending presentations, continuing education, and training programs, and through facilitated learning events. These facilitated learning events can be workshops or tutorials hosted by experts and/or software/hardware vendors, self-directed learning modules, or hackathons. The concept of a hackathon is to assemble a group of interested parties to put focused effort into a specific goal for a short period of time. The amount of time will vary based on the topic at hand and the number of participants available.

In Fall 2019, the Computational Information Sciences and Technology Office (CISTO) at NASA Goddard Space Flight Center partnered with the NASA Goddard Solar Physics Lab to develop a hackweek concept that would help heliophysics scientists expand their knowledge and use of AI/ML in high-performance computing (HPC) systems. We engaged scientists from several disciplines and found external partners in NVIDIA Corporation and the University of Maryland Geographical Sciences department. The external partners were needed to provide a venue that did not require badging and to provide a publicly accessible processing resource accessible to non-credentialed users.

An in-person event was planned for early June 2020 to allow time to advertise, build a planning committee, and develop a program. Due to the coronavirus pandemic, the event was pushed back to late August and converted to a virtual event. Ultimately, there were ~45 applicants from several institutions (domestic and international) who participated for all or part of the event. There were two days of tutorials followed by three days of actual hands-on work (i.e., “hacking”) spread over two weeks between August 20 and August 28, 2020. During the actual hacking, five project mentors led teams of up to eight people to work through specific science initiatives. The resulting code and data products were all stored in GitHub repositories for future use, and at least two of the projects are planning formal presentations or publications based on the work of the hackweek.

Who would like to plan (and replan) a scientific hackweek during a global pandemic and socio-economic crisis?
Apparently, I would! Along with a superb team of planners, sponsors, and supporters—volunteers from across the globe. Without this talented group of people sharing their time and energy, the Heliophysics Hackweek 2020 (HW20) event simply could not have come together, pivoted quickly, and become such a success. Indeed, the combined knowledge, ideas, and experience of the entire Heliophysics Hackweek 2020 planning team, pictured below, resulted in a surprisingly productive and enjoyable event for all 60 participants, despite the many challenges encountered from idea to delivery during the 2020 pandemic.

(Left to right)
Top row: Mark Carroll, Barbara Thompson, Christoph Keller, Burcu Kosar, Brent Smith, Jim Shute, Julia Levites, and Jordan Caraballo Vega.
Middle row: Sean Keefe, Anthony Arendt, Jules Kouatchou, Jack Ma, Chris Bard, Michael Kirk, Jim Hooks, and Andi Moore.
Bottom row: Stan Posey, Zahra Ronaghi, Benoit Tremblay, David Hall, Matthew Penn, Ryan McGranaghan, Matthew Nicely, and Steven Forsyth.

Where did this idea for a NASA hackweek come from?
This hackweek began as a seed of an idea for an event inspired by the open science hackweek model used by the University of Washington (UW) eScience Institute—a model that the planning team followed with generous time and support from our hackweek guru, Anthony Arendt, Senior Research Scientist with the Polar Science Center at the UW Applied Physics Laboratory. It also began as an idea from Mark Carroll of the NASA Goddard Science Task Group (STG) to help NASA scientists practice machine learning in a 5-day, intensive collaboration. It also began as an idea kicked around by Barbara Thompson and her colleagues about holding a heliophysics hackathon.

What brought these grassroots ideas together was interest across NASA Goddard from both heliophysicists and Earth scientists in trying the hackweek approach to use and learn more about AI/ML together. After initial discussions, it was decided to focus on a heliophysics hackweek for the first NASA-sponsored event. (An Earth science hackweek may be in the works—stay tuned.) The University of Maryland and Jack Ma from the Geographical Sciences Department jumped on board quickly with the idea and offered to host the hackweek in one of their smart classrooms specifically designed for media and small-group collaboration. The HW20 planning team grew to include more heliophysicists and data scientists. With ongoing support from Anthony Arendt, we all began learning about the open science model and how to use that planning model as a template for the hackweek. A HW20 GitHub and Wiki was built, the event was advertised, an online application was published, and applications came pouring in from around the globe. The momentum was building.

Then, what happened?
The global COVID-19 pandemic and resulting socio-economic crisis happened. Despite this, and the waning energy and time of busy and tired scientists, the planning team first postponed and then converted the event to a virtual one. Jim Hooks and the NVIDIA team, Goddard computational scientist Jules Kouatchou, and NCCS computer engineer Jordan Caraballo Vega stepped up to provide even more planning support and hands-on training for participants. The SMCE team from CISTO was particularly agile and supportive. Ramon Ramirez Linan and Aaron Skolnik stood up five JupyterHub systems and onboarded approximately 25 people in the span of just a few days—and rebuilt those systems in less than a day after they were inadvertently shut down. It was also handy to have that familiar, collaborative, and highly agile team involved for support as we got closer to the event.

What was surprising to the planning team was that almost all of the applicants stayed the course from March, when the in-person event was first advertised and most people applied, and collaborated virtually with us at the August event. Several applicants pitched their projects in July and became hackweek project mentors, helping form five highly collaborative hackweek project teams. Clearly, there was strong enough interest in collaboration, open science, machine learning, and the virtual hackweek to sustain the momentum of the participants and supporters, even during such a challenging year.

Overall, was the project a success?
Absolutely! Even in the midst of a global pandemic and with two postponements, 60 participants and planners came together to plan, train, learn, code, collaborate, and commit a significant amount of time, energy, knowledge, and creativity to the event, which was a success according to all involved. You can see more photos and read more first-hand feedback in the blogs that follow.

What was your favorite part of the whole hackweek experience?
Meeting and working side-by-side with 60 new brilliant and creative people. Supporting open science and collaboration. And learning, always learning!

What was the hackweek like?
To answer that question, I suggest you read the blogs from the participants. I would like to thank the planners, sponsors, supporters, advisors, project mentors, and participants for helping make this event successful, in the most extraordinary of circumstances. See you next year!


Above: Members of the HW20 training team from NASA and NVIDIA. Clockwise from top left: Steven Forsyth, Jason Tichy, Matthew Nicely, Jim Hooks, Jordan Caraballo-Vega (NASA), David Hall, Zahra Ronaghi, Jim Hooks, and Matthew Penn.

What is your specialty?
I am the Business Development and NASA Account Manager for NVIDIA Corporation, an American technology company incorporated in Delaware that designs graphics processing units (GPUs) for various industries.

How/why did you get involved in the Hackweek?
The NVIDIA team had been collaborating with the NASA Goddard Heliophysics team for quite some time, and we were very excited to be invited by NCCS to sponsor and support the Heliophysics Hackweek 2020 (HW20) event.

What was your role/were your roles in the HW?
I was part of the planning team and responsible for coordinating all NVIDIA support for the event, including GPU compute resources, NVIDIA technical participants, and training sessions.

What did you enjoy and/or learn from/during the event? Are you better versed in AI/ML or other skills now that you participated, and if so, which?
It was a great experience, learning how to successfully pivot a hackweek from in-person to a virtual event. I enjoyed working with the planning committee and learning about the fantastic projects and challenges that the project mentors brought to the hackweek.

What did you learn about using GPUs for AI/ML during the event? What did you learn about NCCS and its GPUs and other high-end computing resources?
HW20 was a great opportunity to help demystify the use of GPUs for AI/ML to dozens of scientists, and also to promote NCCS GPU resources for those that weren’t aware of how to access them.

Did you like the collaborative nature of the hackweek? Advantages/disadvantages?
Hackweek collaboration was critical in bringing data scientists, GPU programming experts, and heliophysics domain scientists together to more quickly solve a scientific challenge, benefiting the entire team via cross-pollination of expertise.

Was the virtual mode of collaboration (as opposed to an in-person event) of the hackweek in/convenient for you? Which platform did you use or like the most: GitHub, Slack, Zoom, or MS Teams?
The virtual event certainly had its challenges versus an in-person event, but by using the combination of all available tools (GitHub, Zoom, Slack, MS Teams), it proved successful in overcoming the shortfalls of any one tool.

Is/Why is collaboration important to you/science?
NASA was an early adopter of using collaboration to help jumpstart AI/ML projects through the NASA Frontier Development Lab Program, which proved how critical it is to pair domain scientists with data science experts in order to speed the development of AI-based scientific breakthroughs, bridging the gap in knowledge that is often too difficult to overcome otherwise.

What was the hackweek’s overall impact to you personally, (to your current/future work in AI/ML) and/or to your hackweek team?
This event was impactful in jump-starting fruitful collaborations between NASA Heliophysics scientists, experienced data scientists, and GPU programming experts, laying a foundation for continued work in applying AI/ML techniques toward the next generation of potential breakthroughs in heliophysics related research.


Above: HW20 planners and trainers. Clockwise from top left: Jules Kouatchou, Mark Carroll, Zahra Ronaghi (NVIDIA), Jordan Caraballo-Vega, Matthew Nicely (NVIDIA), Jules Kouatchou, and Sean Keefe.

What is your scientific discipline/specialty?
I am a Computational Scientist at NASA Goddard Space Flight Center. I earned a Master’s degree in electrical engineering at Télécom Paris and a Ph.D. in computational and applied mathematics at George Washington University. I have over a decade of experience in porting and implementing numerical atmospheric models (such as MM5, Eta, CCM3, GMI, GEOS-5) on high-performance computers.

How/why did you get involved in the Hackweek?
I was an early and continuing member of the HW20 planning team. I also developed several online, pre-event Python training modules published to the official HW20 GitHub Wiki and conducted live training in Python Data Analysis Library with Scikit-learn for all 60 participants during the first week of the hackweek, coordinating closely with the planning team and NVIDIA trainers.

What did you enjoy and/or learn from/during the event? Are you better versed in AI/ML or other skills now that you participated, and if so, which?
I enjoyed attending the HW20 NVIDIA lectures. They opened my eyes to areas of programming I was not familiar with.

Was the virtual mode of collaboration (as opposed to an in-person event) of the Hackweek in/convenient for you? Which platform did you use or like the most: GitHub, Slack, Zoom, or MS Teams?
Though it was a challenge to have virtual event, it has some advantages too. For instance, members of one group needed assistance in Python, and they were to quickly reach to me. We shared screens and resolved the problems.

Is/Why is collaboration important to you/science?
Working in isolation only allows an approach to data science problems from a narrow point of view. Collaborations open our eyes to the perspectives of others, which often leads to more robust and effective solutions.

Above: Members and supporters of the HW20 "Coronal Holes" Team. Clockwise from top left: Michael Kirk, Silvina Guidoni, Ajay Kumar Tiwari, Matthew Penn (NVIDIA), Andong Hu, Michael Kirk, Tanmoy Samanta, James Staeben, and Benoit Tremblay.

A little background: I am a research scientist (solar physicist) in the Heliophysics Science Division at NASA’s Goddard Space Flight Center. I support the Solar Dynamics Observatory satellite and study the physics of the Sun, the causes of solar variability, and its impacts on Earth. I specialize in extracting science from large numbers of solar images. Beyond my own research efforts, I’m currently helping launch of the Center for HelioAnalytics (CfHA) at NASA Goddard.

How/why did you get involved in the Hackweek?
I got involved with Helio Hackweek 2020 to jump-start innovative science in heliophysics through high-risk and high-reward efforts involving machine learning.

What was your role/were your roles in the HW?
I was involved in some of the early organization for the event with the planning team, and later served as an advisor to the Coronal Holes project team. I also provided the team with images for the project as well as specialized code for working with solar images.

What did you enjoy and/or learn from/during the event? Are you better versed in AI/ML or other skills now that you participated, and if so, which?
I am better able to move my workflow onto a cloud computing platform through the guidance of NVIDIA and NCCS.

What did you learn about using GPUs for AI/ML during the event? What did you learn about NCCS and its GPUs and other high-end computing resources?
The ability of NCCS to troubleshoot problems in real time was impressive to me: that kind of interaction gives me the confidence to go back to difficult computational problems in the future.

Did you like the collaborative nature of the hackweek? Advantages/disadvantages?
I am impressed with the international collaboration we were able to generate over video chat, despite being separated by so many time zones.

Was the virtual mode of collaboration (as opposed to an in-person event) of the Hackweek in/convenient for you? Which platform did you use or like the most: GitHub, Slack, Zoom, or MS Teams?
Slack, GitHub, and Zoom were absolutely essential. We used Zoom breakout rooms, running most hours of the day, and constantly used chat on our Slack channel.

What was the hackweek’s overall impact to you personally, (to your current/future work in AI/ML) and/or to your hackweek team?
This hackweek provided me with a concrete example of how to do intensive and robust collaboration with colleagues from numerous time zones and produce tangible results in a virtual setting.

Members and supporters of the HW20 "Accelerating AIApy" Team.
Left: Chris Bard:
(From left to right) Top Row: Luiz Guedes dos Santos, Paul Wright, Zahra Ronaghi, and Will Barnes.
Middle Row: Raphael Attie, Jack Ireland, Chris Bard, and Matthew Nicely.
Bottom Row: Mark Cheung, Matthew Nicely, Jinwoong Yoo, and Peter Schuck.

A little background: I am a research astrophysicist in the Heliophysics Science Division at NASA Goddard Space Flight Center.

What is your scientific discipline/specialty?
I mainly write code for researching space plasmas, running both large-scale simulations of planetary magnetospheres and small-scale environmental reconstruction from spacecraft data using neural networks.

What is your affiliation (university, agency, company)
I work in the Heliophysics Science Division’s Geospace Lab at NASA Goddard.

How/why did you get involved in the Hackweek?
I thought it was a great opportunity to get to know other people and learn science and computer skills from one another. Also, coding is fun—and more fun with other people!

What was your role/were your roles in the HW?
I was one of the original members of the planning team, and I later became a hackweek project team mentor.

What did you enjoy and/or learn from/during the event? Are you better versed in AI/ML or other skills now that you participated, and if so, which?
I learned a lot about open-source collaboration and using GPUs with Python (via CuPy, Numba).

What did you learn about using GPUs for AI/ML during the event? What did you learn about NCCS and its GPUs and other high-end computing resources?
I have already been using NCCS ADAPT for some of my simulations.

Did you like the collaborative nature of the hackweek? Advantages/disadvantages?
Yes, the hackweek did a great job of bringing people together, especially from different parts of the world. It was sometimes tricky, since we could only have one conversation at a time through the MS Teams chat room. It would have helped to have multiple topics spun off into their own breakout rooms.

Was the virtual mode of collaboration (as opposed to an in-person event) of the Hackweek in/convenient for you? Which platform did you use or like the most: GitHub, Slack, Zoom, or MS Teams?
Slack was the best virtual collaboration tool for my team. I would say the challenge for my project was that we did not have a way to simultaneously edit and look at code, [since it became a virtual event], so we had to continuously upload code (or paste snippets to Slack) throughout the hackweek while working on our local versions.

Is/Why is open source software development important to science?
Open source codes are important to science because they allow for easier reproducibility of scientific results and analysis; they provide a good foundation for other scientists to more easily build upon previously-established code and analysis, without having to write their own code from scratch.

What was the hackweek’s overall impact to you personally, (to your current/future work in AI/ML) and/or to your hackweek team?
Working on my hackweek project introduced me to several people whom I would never have met otherwise, and it gave me a chance to work with people who don’t work in my scientific discipline.


Above: Members and supporters of the HW20 "MMS Data Hunt" Team. Clockwise from top left: Nick Dutton, Jack Payette, Rachel Bailey, Brent Smith, Ajesh Pillai, Laura Cutrer, Bhaskar Bishnoi, Brent Smith, and Space Mascot. (Not pictured: Calley Tinsman.)

A little background: I am a Senior Scientific Developer at Johns Hopkins University's Applied Physics Laboratory (JHU APL).

What is your scientific discipline/specialty?
My background was primarily in theoretical heliophysics concerning shock theory of solar energetic particles (SEPs). However, for the past decade, my experience has been within the field of Earth science, primarily focused on Python programming on HPC systems and web-based applications.

How/why did you get involved in the Hackweek?
Originally, when I was still at NASA Goddard, I got involved in the hackweek project as a planning team member and potential event Python trainer, since I had a strong Python programming background and a working knowledge of heliophysics. After moving to the JHU APL, I was offered the role as a project mentor. Throughout all of these changes, I was eager to be a part of this event, because I have learned that there is a lot of community-based research to be accomplished in a venue such as a hackweek. I saw the benefit of this type of event from prior experience in several PyCon conferences in which open source project developers were ready with projects that participants with any level of Python experience could contribute.

What was your role/were your roles in the HW?
I was one of the original HW20 planners and ultimately became the project mentor for the “MMS Data Hunt” team.

What did you enjoy and/or learn from/during the event? Are you better versed in AI/ML or other skills now that you participated, and if so, which?
I enjoyed the experience of leading a diverse group of individuals that, when combined, can truly act as a research group with a common goal.

What did you learn about using GPUs for AI/ML during the event? What did you learn about NCCS and its GPUs and other high-end computing resources?
I was already knowledgeable about GPUs and AI/ML. I would have learned more if we had had more time working on the projects in question. [Editor’s note: Time is normally limited in a hackweek, and due to the COVID-19 pandemic, this event was replanned from a five-day, in-person event to a virtual, partial-day event spread over two weeks.]

Did you like the collaborative nature of the hackweek? Advantages/disadvantages?
The advantage of this event is that we were able to pull from the experience and knowledge of individuals that we may not already possess. The disadvantage was primarily from the viewpoint of those that did not know or have experience with programming in general. My time was divided between helping those who were primarily beginners to understand what was going on and trying to keep those who were more advanced energized about the project.

Was the virtual mode of collaboration (as opposed to an in-person event) of the Hackweek in/convenient for you? Which platform did you use or like the most: GitHub, Slack, Zoom, or MS Teams?
Due to the [COVID-19 pandemic and] virtual nature of the event. I feel that we could have more easily got the team on the same page faster if we had held the event in person. Our team used GitHub for project and code development, Slack for announcements and side discussions, and Zoom for team meetings.

Is/Why is open source software development important to science?
Open source software development is immensely important to research within the sciences, as many industries rely upon these free utilities to perform a majority of their work. It is also important that those groups/individuals that use these community-created platforms and libraries contribute back to the community of open source software to close the loop and give back to those truly making great strides in digital capabilities. I challenge readers to learn more about why contributing to open source software is important: https://opensourcefriday.com.

What was the hackweek’s overall impact to you personally, (to your current/future work in AI/ML) and/or to your hackweek team?
The Helio HackWeek provided a platform for myself and others to experience and contribute to an open source solution to a research-based goal or problem. It is a platform that many other industries should utilize for creativity, quick advancements, and team-building in general.

Above: Members and supporters of the HW20 "Coronal Holes" Team. Clockwise from top left: Silvina Guidoni (twice), Ajay Kumar Tiwari, Matthew Penn (NVIDIA), Andong Hu, Michael Kirk, Tanmoy Samanta, James Staeben, and Benoit Tremblay.

A little background: I am a heliophysicist who focuses on the study of solar eruptions. I am particularly interested in physical processes of reconnected plasma, such as magnetic reconnection and particle acceleration. As an Assistant Professor in Physics at American University, I teach undergraduate level courses and mentor students on research projects related to heliophysics. I have been a member of NASA and NSF proposal review panels and served as a peer reviewer for several professional journals. When I’m not thinking about equations, programming, or the vastness of space, I practice T'ai chi, an activity I have enjoyed from more than two decades.

What is your scientific discipline/specialty?
I am a heliophysicist, and I model physical processes in solar eruptions.

What is your affiliation (university, agency, company)?
I am a Professor at American University, Washington, DC.

How/why did you get involved in the Hackweek?
I am interested in using machine learning (ML) and GPUs for my research and try to get my hands on everything that will get me closer to that goal, so I immediately applied when I saw the email advertising the Hackweek.

What was your role/were your roles in the HW?
I am a beginner in ML and GPUs, so I mostly provided expertise in data preparation, but I learned a lot about ML and GPUs during the Hackweek because my teammates took the time to explain everything that was being done. My teammates rocked!

What did you enjoy and/or learn from/during the event? Are you better versed in AI/ML or other skills now that you participated, and if so, which?
I enjoyed sharing a common goal with a diverse group of people; each person contributed with their expertise and enthusiasm.

What did you learn about using GPUs for AI/ML during the event? What did you learn about NCCS and its GPUs and other high-end computing resources?
I learned that using GPUs is easier than I expected and more accessible.

Did you like the collaborative nature of the hackweek? Advantages/disadvantages?
That was the best part.

Was the virtual mode of collaboration (as opposed to an in-person event) of the Hackweek in/convenient for you? Which platform did you use or like the most: GitHub, Slack, Zoom, or MS Teams?
It was convenient because I did not have to travel. We mostly used GitHub and Zoom.

Why is collaboration important to you/science?
Because “the whole is greater than the sum of its parts.”

What was the hackweek’s overall impact to you personally, (to your current/future work in AI/ML) and/or to your hackweek team?
The hackweek showed me how convenient and easy is to use GPUs and ML, so I am planning to develop a program to simulate magnetic field evolution in solar flares that uses GPUs and analyze its result using ML techniques. The event was very successful. My team was really good. I am so grateful for having been able to attend the Hackweek!


Above: Members and supporters of the HW20 "Accelerating AIApy" Team. Clockwise from top left: Yihua Zheng (twice), Lutz Rastaetter, Sheng Huang, Jian Li, Liang Wang, and David Hall (not pictured: Ryan McGranaghan and Sean Keefe).

A little background: I am a Research Astrophysicist at NASA Goddard Space Flight Center. I have been involved in providing and improving space weather services to NASA robotic missions, including planetary missions. One of my current interests is to expand my research domain by analyzing space weather events from a broader perspective—starting with the originator, the Sun—to understand the full chain of physical processes involved in space weather events and their effects throughout the interplanetary space, combining modeling results and observations.

What is your scientific discipline/specialty?
The science of space weather and Earth’s magnetosphere.

How/why did you get involved in the Hackweek?
I heard it from NASA internal email distributions and became interested in attending.

What was your role/were your roles in the HW?
I was a beginner who was ready to learn and to absorb things, working as a member of Liang Wang’s team.

What did you enjoy and/or learn from/during the event? Are you better versed in AI/ML or other skills now that you participated, and if so, which?
We had a great and collegial team, and I learned a lot from each teammate. I know more about artificial intelligence and machine learning (AI/ML) than before and learned how to use Tensorflow and PyTorch in Python.

What did you learn about using GPUs for AI/ML during the event? What did you learn about NCCS and its GPUs and other high-end computing resources?
I learned about CUDA/CuPy GPU accelerated computing. Sean showed us about NCCS services, cool news stories, the high-performance computing platform Discover, and other services.

Did you like the collaborative nature of the hackweek? Advantages/disadvantages?
I enjoyed the collaborative nature of the hackweek and especially our team. There are a lot of advantages: one of which is to expand the collaboration circle. The only disadvantage lies in myself. I could have come to the hackweek with a better background in Python programming and the science and math behind all of the AI/ML algorithms.

Was the virtual mode of collaboration (as opposed to an in-person event) of the Hackweek in/convenient for you? Which platform did you use or like the most: GitHub, Slack, Zoom, or MS Teams?
We used GitHub, Slack and MS Teams. Knowingly, MS Teams had some problems and limitations, but it didn’t cause much hindrance.

Is/Why is AI/ML important to you/science?
AI/ML has seen its growth in space weather forecasting and for understanding complex systems. I would like to understand more and hopefully to apply it to some aspects of my research, such as forecasting radiation belt electrons (killer electrons) and the intensity of solar energetic particles (that pose radiation hazards), etc.

What was the hackweek’s overall impact to you personally, (to your current/future work in AI/ML) and/or to your hackweek team?
I enjoyed learning from and working with a bunch of friendly, smart and highly efficient team members. In addition, I cherish great resources from my team and other teams, which I can go back to later.


Above: Members and supporters of the HW20 "Atmospheric Relativistic Electron Precipitation Predictions" Team. Clockwise from top left: Sapna Shekhar, Mike Shumko, Sapna Shekhar, Luisa Capannolo, Aziz Diaby, Steven Forsyth (NVIDIA), Jian Li, David Linko, and Barbara Thompson.

A little background: My research involves exploration of space plasma that affects satellites, particularly in the radiation belt regions where high-energy particles from the Sun are known to cause satellite damage through external or internal charging and satellite drag.

What is your scientific discipline/specialty?
My research mostly spans magnetospheric space plasma physics.

What is your affiliation (university, agency, company)?
I was a Postdoctoral Research Fellow at Auburn University in Alabama when I applied to the hackweek, and recently transitioned to a new position as a Postdoctoral Researcher at the University of Iowa.

How/why did you get involved in the Hackweek?
I wanted to collaborate on my idea with experts from other fields.

What was your role/were your roles in the HW?
I was a project mentor, and my responsibilities were to keep my team members engaged, assimilate results, and plan the tasks ahead.

What did you enjoy and/or learn from/during the event? Are you better versed in AI/ML or other skills now that you participated, and if so, which?
I really enjoyed interacting with new team members and learned a great deal about feature importance evaluations and data conditioning.

What did you learn about using GPUs for AI/ML during the event? What did you learn about NCCS and its GPUs and other high-end computing resources?
I learned a great deal about the tools and Python libraries available to accelerate computations with GPUs.

Was the virtual mode of collaboration (as opposed to an in-person event) of the Hackweek in/convenient for you? Which platform did you use or like the most: GitHub, Slack, Zoom, or MS Teams?
An in-person event would have surely increased our efficiency, but we did get a good amount of work done at the virtual event. I certainly prefer in-person events, but clearly that wouldn’t have been possible [due to the COVID-19 pandemic.] We mostly used GitHub for code sharing.

What was the hackweek’s overall impact to you personally, (to your current/future work in AI/ML) and/or to your hackweek team?
I have learned that the space research community needs to tap the full potential of data from satellites to explore space weather, and AI/ML allows us to do that.


Above: Members and supporters of the HW20 "Accelerating AIApy" Team. Clockwise from top left: Liang Wang, Yihua Zheng, Lutz Rastaetter, Sheng Huang, Sean Keefe, Ryan McGranaghan, Jian Li, Liang Wang, and David Hall (NVIDIA).

What is your scientific discipline/specialty?
Space plasma physics/Magnetosphere modeling.

What is your affiliation (university, agency, company)
Princeton University

How/why did you get involved in the Hackweek?
I heard about Heliophysics Hackweek 2020 at a conference. I would like to use machine learning (ML) in my research, and I found that self-education in ML during my free time was not sufficient.

What was your role/were your roles in the HW?
Team lead.

What did you enjoy and/or learn from/during the event? Are you better versed in AI/ML or other skills now that you participated, and if so, which?
I enjoyed full interaction with the project team in almost every stage of our project. I consolidated my knowledge in the fundamentals of ML and wrote real codes with NVIDIA experts and team members. I think the entire team learned together and made progress together, which again encouraged everyone to participate and contribute. I also learned GPU techniques like CuPy and Numba on the GPU, which I now use in my research.

What did you learn about using GPUs for AI/ML during the event? What did you learn about NCCS and its GPUs and other high-end computing resources?
I learned to use CuPy and Numba-GPU as well as how to use GPU in Tensorflow and PyTorch. I also learned the basic workflow of the NVIDIA on-demand system, as well as the RAPIDS suite of open source software libraries and APIs. I learned about the availability of NCCS resources for NASA-funded researchers, which I plan to take advantage of. I do hope some of the workflow I learned during the hackweek could be available in a similar mechanism on NCCS, if I have the opportunity to use them.

Did you like the collaborative nature of the hackweek? Advantages/disadvantages?
I highly appreciate the collaborative nature of the hackweek. I feel a friendly setup encouraged me to comfortably apply ML to my research and understand other people’s research as well. I benefited from this hackweek in both coding skills and thinking, since we have different, but related backgrounds.

I do think having a balanced team is important for such collaboration to be productive/helpful. For this reason, one disadvantage of a highly collaborative session is that not everyone is prepared (since everyone is busy has their own work to do). I can imagine a person entirely new to the subjects could struggle if insufficient guidance is provided from the lead and from the organizers.

Was the virtual mode of collaboration (as opposed to an in-person event) of the Hackweek in/convenient for you? Which platform did you use or like the most: GitHub, Slack, Zoom, or MS TEAMS?
For a future hackweek/hackathon, I might even prefer a virtual one due to the reduced stress of travel and funding, yet relatively small loss on the efficiency. I also feel virtual events could encourage more people to join, though that also sets higher standards on the online content planning. I use GitHub, Slack, and Zoom on a daily basis, so I prefer those platforms over their alternatives. I like GitHub the most for its organization of code and projects.

Is/Why is AI/ML important to you/science?
AI/ML seems promising to improve physics-driven modeling at a fundamental level, and it is complementary to traditional approaches like tedious theoretical derivation and formidable algorithm development.

Is/Why is open source software dev’t. important to science?
I find open-source software extremely important for science, although it has unfortunately been underestimated in general due to historical reasons (particularly for heliophysical sciences) and funding competitions. Open source helps to create collaborative communities that bring together scientists with different skill sets and scientific objectives. Open source software also requires that developers with a scientific background write more readable and sustainable codes and reduce errors. I think science, including fundamental sciences, has entered an era where open source is mandatory for deeper involvement of the entire community, not only open source “heroes.”

Is/Why is collaboration important to you/science?
Space sciences are intrinsically multidisciplinary, and they are becoming more so. Take AI/ML for example: scientists would love to take advantage of the newest developments, but working in solo mode is a bottleneck, since many career scientists do not have the time and/or funding to explore the newest tools and knowledge.

What was the hackweek’s overall impact to you personally, (to your current/future work in AI/ML) and/or to your hackweek team?
The biggest impact to myself and, I believe, to everyone on our team, is that we got over the hurdle between tutorial content on ML and actually applying them in our current and future research.


Above: Members and supporters of the HW20 "Coronal Holes" Team. Clockwise from top left: Ajay Kumar Tiwari, Silvina Guidoni, Ajay Kumar Tiwari, Matthew Penn (NVIDIA), Andong Hu, Michael Kirk, Tanmoy Samanta, James Staeben, and Benoit Tremblay.


A little background: I’m from the eastern part of India (Kolkata). I started off as a microbiologist working on bacterial communication before I was drawn to astrophysics. I love to travel and have trekked into the Sahayadri (the Western Ghats).

What is your scientific discipline/specialty?
My specialty is in solar physics/space weather.

What is your current affiliation (university, agency, company)
I am currently a postdoctoral researcher at Centrum Wiskunde and Informatica (CWI), the national research institute for mathematics and computer science in Amsterdam, Netherlands.

How/why did you get involved in the Helio Hackweek?
I have always enjoyed hackweeks, so participating in a hackweek for heliophysics made perfect sense.

What were your roles in the Hackweek?
I pitched a project idea to identify polar coronal holes, combining multi-stereoscopic observations of the Sun (SOHO, STEREO-A, STEREO-B). This idea was approved, and our Coronal Holes team worked on this project together at the hackweek with great success.

What did you enjoy and/or learn from/during the event? Are you better versed in AI/ML or other skills now that you participated, and if so, which?
I most enjoyed interacting with people and learning from them, at the same time leading a project from a concept to proof of concept in a week. It was good to go over the fundamentals of ML during the workshop tutorials. I am better versed in understanding the architecture of neural networks. This experience is going to be quite valuable in my future.

What did you learn about using GPUs for AI/ML during the event? What did you learn about NCCS and its GPUs and other high-end computing resources?
I went in knowing almost nothing about GPUs. The tutorials helped me understand that using GPUs translates to speeding up of some of my scripts, so I will be using GPUs in my research.

Did you like the collaborative nature of the hackweek? Advantages/disadvantages?
I very much enjoyed the collaborative nature of the Helio Hackweek. It was challenging in the beginning, but it became quite interesting after a while, and I got to learn so much from my peers

Was the virtual mode of collaboration (as opposed to an in-person event) of the Hackweek in/convenient for you? Which platform did you use or like the most: GitHub, Slack, Zoom, or MS Teams?
The virtual mode indeed needed some getting used to, but once we got going, it was a good experience. I used Zoom and Slack quite often, along with GitHub.

Why is AI/ML important to you/science?
The volume of data that we already have access to in solar physics is enormous, and with the new generation of high-resolution instruments, the data is going to get even more difficult to deal with. Machine learning becomes one of the best tools to use this large volume of data to work on interesting problems.

What was the hackweek’s overall impact to you personally, (to your current/future work in AI/ML) and/or to your hackweek team?
The Helio Hackweek collaboration was a success. We had fun and learned a lot as well. I gained some new friends, and we are keeping the collaboration alive. Our team plans to submit the results that we obtained during the hackweek to a conference. All in all, it was a huge success for me. Let’s meet at HW21!



Related Links

Sean Keefe, NASA Goddard Space Flight Center