Equity, Scalability, and Sustainability of Data Science Infrastructure

Suen, Anthony; Norén, Laura; Liang, Alan; Tu, Andrea

doi:10.25080/majora-4af1f417-002

Cited by 4 publications

(4 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A second obstacle integrating software tools into scientific practice is that software-based learning requires additional education infrastructure. [SNLT18] document the challenges in providing JupyterHub with automatic grading extensions at universities and colleges; they find that many institutions do not have the resources or deep IT expertise necessary to build and maintain this infrastructure. The growing necessity of cloud-based computational notebooks for assignments and exploration in scientific education therefore raises concerns about social equity.…”

Section: Computing In Education and Sciencementioning

confidence: 99%

Proceedings of the 19th Python in Science Conference

Varoquaux¹,

Walt²,

Millman

2020

Proceedings of the Python in Science Conference

View full text Add to dashboard Cite

SciPy 2020, the 19th annual Python in Science Conference, was held July 6-12, virtually via the conference platform Crowdcast. Due to the COVID-19 pandemic, the SciPy conference was held online. The SciPy Conference brings together a community of researchers, engineers, and programmers dedicated to the advancement of scientific computing through open source Python software.The two main conference themes for 2020 were high performance computing; and, machine learning and data science. Discipline-specific symposia included astronomy and astrophysics; biology and bioinformatics; materials science; earth, ocean, geology, and atmospheric science; and a new symposium dedicated to fostering conversations among maintainers of the open source infrastructure that help power the worlds of scientific discovery and engineering. As was the case in 2019, there were plenary sessions for updates about key scientific software libraries, and three sessions of the ever-popular lightning talks, which this year included SciPy's youngest speaker, Artash Nath, discussing machine learning approaches in exoplanet research.The first conference keynote lecture was delivered by Anne Carpenter, who discussed the history of CellProfiler in the context of developing academic software, current application of the scientific software stack to problems in biology, and future directions for tasks like drug discovery, powered by machine learning. Andrew Chael delivered the second keynote, about the large, interorganizational effort to take the first photograph of black hole M87, and the role of scientific software in that project. This year's diversity plenary was given by Bonny McClain, who delivered an interactive lecture about bias in data, and how to think about measuring what people haven't thought about measuring before.The online format permitted a larger-than-usual number of participants, ultimately attracting 1412 participants from a recordbreaking 57 countries, making this the largest SciPy Conference yet. Participants reported that they enjoyed the broader access to beginner tutorials for popular libraries like PyTorch and xarray -something that would not be possible without having the conference at least partially online. Birds of a Feather (BoF) sessions were organized around the topics of packaging, diversity, Python in education, hardware, and SciPy 2021 with great attendance due to the online format. Sprints that usually gather around tables in conference rooms took the conversation to virtual tables using a ¶ Oracle § Mozilla

show abstract

Section: Computing In Education and Sciencementioning

confidence: 99%

Proceedings of the 19th Python in Science Conference

Varoquaux¹,

Walt²,

Millman

2020

Proceedings of the Python in Science Conference

View full text Add to dashboard Cite

show abstract

“…Data 8 utilizes ok.py, a Berkeley developed solution that has a plethora of features for large and diverse computer science and data science classes. However, this comes with a complexity cost for instructors who only need a subset of these features and sysadmins operating an okpy server installation [Suen18]. On the other hand, Data 100, the upper division core data science course, utilizes nbgrader, an open source grading solution built for Jupyter Notebooks.…”

Section: Setting Campus Wide Educational Cyber-infrastructurementioning

confidence: 99%

Accelerating the Advancement of Data Science Education

Dusen¹,

Suen²,

Liang³

et al. 2019

Proceedings of the Python in Science Conference

Self Cite

View full text Add to dashboard Cite

We outline a synthesis of strategies created in collaboration with 35+ colleges and universities on how to advance undergraduate data science education on a national scale. The four core pillars of this strategy include the integration of data science education across all domains, establishing adoptable and scalable cyberinfrastructure, applying data science to non-traditional domains, and incorporating ethical content into data science curricula. The paper analyzes UC Berkeley's method of accelerating the national advancement of data science education in undergraduate institutions and examines the recent innovations in autograders for assignments which helps scale such programs. The conversation of ethical practices with data science are key to mitigate social issues arising from computing, such as incorporating anti-bias algorithms. Following these steps will form the basis of a scalable data science education system that prepares undergraduate students with analytical skills for a datacentric world.

show abstract

“…2 While server-based solutions have some costs associated, for students with internet access they can increase equity of education as all students have equal computational power regardless of their own computing hardware. 1,3 These technology challenges are magnified for those teaching data science focused Massively Open Online Courses (MOOCs). MOOCs are typically offered to thousands of learners across the globe completely asynchronously, increasing the number of unique computing environments and reducing instructor contact for individual-level support.…”

Section: Introductionmentioning

confidence: 99%

“…4,5 Others use technology embedded in the platform's learning management system (e.g., shared JuptyerHub). 3,6 Importantly, these data science MOOCs have not had a particular domain focus and thus can use openly available or non-sensitive data in their courses.…”

Section: Introductionmentioning

confidence: 99%

Developing and Deploying a Scalable Computing Platform to Support MOOC Education in Clinical Data Science

Mayer

Russell

Wilson

et al. 2020

Preprint

View full text Add to dashboard Cite

One of the challenges of teaching applied data science courses is managing individual students’ local computing environment. This is especially challenging when teaching massively open online courses (MOOCs) where students come from across the globe and have a variety of access to and types of computing systems. There are additional challenges with using sensitive health information for clinical data science education. Here we describe the development and performance of a computing platform developed to support a series of MOOCs in clinical data science. This platform was designed to restrict and log all access to health datasets while also being scalable, accessible, secure, privacy preserving, and easy to access. Over the 19 months the platform has been live it has supported the computation of more than 2300 students from 101 countries.

show abstract

Equity, Scalability, and Sustainability of Data Science Infrastructure

Cited by 4 publications

References 0 publications

Proceedings of the 19th Python in Science Conference

Proceedings of the 19th Python in Science Conference

Accelerating the Advancement of Data Science Education

Developing and Deploying a Scalable Computing Platform to Support MOOC Education in Clinical Data Science

Contact Info

Product

Resources

About