ADSA Annual Meeting - 2024

There is a growing awareness about the ethical and societal implications related to the use of data in AI and Generative AI. It is therefore critical to integrate Ethics in the curriculum of students in Data Science and Computer Science. As future practitioners, students need to be equipped with the language and critical thinking skills that enable them to engage in the discussions around the inherent normativity in building socio-technical systems. However, there are serious challenges in the implementation of normative elements in the curriculum of Data Science and Computer Science, namely the contrasting nature of these disciplines, poor collective knowledge and best practices, and lack of support and training for instructors in technical fields who wish to engage in these discussions. The session on Methods for Teaching Ethics in Data Science aims to foster a reflection on the challenges associated with teaching Ethics in technical or quantitative domains, sharing experiences and progress made, and establishing core best practices. Our goal is to build a community empowering data science instructors in teaching the new generation of data scientists and tech practitioners. Building on the first workshop on methods for teaching ethics in data science which tool place in May 2023, we propose the second edition of the workshop with one 30 minute plenary talk (by Dr. Benedetta Giovanola, Professor, University of Macerata) followed by 10-15 min talks on teaching AI Ethics selected by a technical program committee (see draft CFP below). We will publish the proceedings online through ADSA.

Higher education institutions need to produce individuals with data science acumen to meet the demands of a 21st century data science workforce. While most organizations and agencies have access to data, many cannot capitalize and benefit from it, as they lack access to staff with interdisciplinary skillsets. Higher education institutions tend to operate in disciplinary silos and often prepare a workforce that is accustomed to sourcing solutions for specific disciplinary problems as opposed to interdisciplinary problems relevant to current technological and societal contexts. Furthermore, the best talents from data science programs often choose to work for large technology-oriented companies. Thus, it is very difficult for small, mid-sized companies and public sector organizations to gain access to graduates with data science skills. Several universities across the U.S. created Data Science for Social Good (DSSG) programs to address the data science talent needs of public sector organizations. However, the impacts of DSSG programs on student experiences and workforce demands are understudied. Florida DSSG program has been conducting studies on student experiences and how those experiences lead to skillsets that match workforce demands. During the session, we will present our study findings and conduct live reflection activities using the Riff AI tool to gather ADSA participants’ perspectives on the data science workforce, and STEM student educational and co-curricular programs.

Radical changes in platform and regulatory policy have impacted how researchers collect, analyze, and share social media data. Elon Musk bought Twitter and shut down the academic API. Reddit curtailed large-scale access to its data and then went public. Meta sunset CrowdTangle. The European Union demanded that platforms provide researchers access to data. And yet, social media remains an invaluable resource for science. In this session, we will discuss the legal, ethical, and computation challenges social media data collection, sharing, analysis, and archiving pose. Our speakers/panelists include researchers who have used authorized APIs and other data collection approaches, staff from the Social Media Archive at ICPSR who support users in sharing and reusing data, and platform insiders who are responsible for complying with data sharing regulations.

Container-driven Reproducible Research Made Simple
​Ronaldas Lencevicius (University of California, Santa Barbara)
Scholarship in data science should consists of a complete software development environment along with instructions for all the results and figures. However, fully specifying and reproducing an arbitrary data science workflow can often be challenging, especially with the increasing complexity of software dependencies and computational infrastructure. Furthermore, reproducibility that relies on documentation or language specific tools can involve specialized adjustments and tweaking that many researchers may not have the time or background for. To address this common deficiency, we introduce a computational research framework to the data science community that can specify complex computational environments using an OS-level virtualization technology called containers. We show that the container-driven reproducibility approach balances flexibility and ease of use through Visual Studio Code, a popular code editor. In addition, to alleviate the steep initial learning curve of containers, we introduce a code-generating template repository for further simplifying the initial setup of Python and/or R-based workflows commonly used in data science.

Lightning Talks:

1. Data Science + Environmental Justice = Experiential Nugget on Air Pollution - Ravanasamudram Uma (North Carolina Central University)
2. Generative AI for Social Science Data Archiving at ICPSR - Murali Mani (University of Michigan Flint)
3. Data Analytics Program at Washington State University - Sergey Lapin (Washington State University Everett)
4. Non-Invasive and Explainable Assessment of Burn Depth for Surgical Decision-Making Using a Vision-Language Model and Ultrasound Imaging - Md Masudur Rahman (Purdue University)
5. Administrative checkpoints, burdens and human-centered design: Increasing interview access to raise SNAP participation - Jae Yeon Kim (Johns Hopkins University)
6. Assessing the Impact of Large Language Models on Learning Data Visualization Libraries - Saara Uthmaan (University of Washington)

Scientific software is at the heart of data-intensive research projects in nearly every domain. With the increasing availability of generative AI tools like GitHub Copilot and ChatGPT, the practice of programming for research is sure to be affected. Because scientific conclusions frequently depend on code for data analysis, collection, visualization and simulation, the potential impacts of these tools on data-intensive research may be substantial. In this structured discussion, we'll together address the question: How does the use of generative AI code tools change the work involved in writing, using, and maintaining code for data-intensive science? We'll explore how the skills needed to develop, maintain and use high-quality scientific code may be changing. We'll discuss the potential impacts of increased reliance on generative AI code tools on the validity, correctness, and maintainability of scientific code. Finally, we'll also brainstorm what kind of training, practices, and tooling might help scientists best take advantage of generative AI software tools while mitigating risks.

Interdisciplinary research institutes have been playing significant roles in academic research. Data science and AI institutes, however, often face unique challenges because of the nature of data science and AI - they are not only fields of inquiry, but are also becoming essential methodologies for all research fields. As institutions grapple with this reality, they are setting up data science and AI institutes in many different ways, with different charges, funding models, operation models, and relationships with the rest of the campus. A group of leaders of data science and AI institutes at research universities have carried out a study to understand how to best position such institutes on campus, how these institutes can add the most value to the universities, and the sustainability strategies of these institutes. In this session, we will present our findings and invite audience members to share their thoughts, all for the purpose of maximizing the impact of such institutes in academic research.

This session aims to share best practices for upskilling students from diverse majors in using data science to address problems within their respective disciplines. A significant challenge in this endeavor is achieving proficiency in programming-based tools, which often act as a barrier to student success. Non-programming tools, while more accessible and user-friendly for beginners, have inherent limitations such as the size of the datasets they can handle, limiting their effectiveness for more complex analyses. This session will explore the best practices for creating a data analytic pathway that transitions from no-code to low-code to high-code tools. The co-chairs will share their experiences in upskilling students from non-computing disciplines, beginning with spreadsheets, CODAP, and Weka, before transitioning to provide a gentle introduction to accessible high-code tools. The session will include audience participation through table discussions and share-outs, focusing on the following prompts:
1. What tools have you used, and what limitations or obstacles have your students encountered with those tools?
2. What resources do you believe are necessary or helpful to enable your students to overcome these obstacles? (Basically, your wish list.)
3. What ideas have you implemented, or do you have, to bring a wider group of students, teachers, and others into data science?
This work is supported through NSF grant #2245958.

The session will feature leaders in AI/ML methods for space weather forecasting, demonstrating the promises and opportunities for space science and AI researchers.

Session Chair: Lulu Zhao (University of Michigan - Climate and Space Sciences and Engineering)

1. Data Quality Issues in Flare Prediction Using Machine Learning Models - Ke Hu (University of Michigan)
2. Predicting Solar Energetic Particle Events Using Machine Learning Algorithms with Flare Features - Chia-Yun Li (University of Michigan)
3. An Overview of Surrogate Models for Synthetic White Light Images in the Space Weather Modeling Framework - Aniket Jivani (University of Michigan)
4. Regression Estimate Recalibration using Kernelized Stein Discrepancy Scores: Applications in Space Weather - Matthew McAnear (University of Michigan)
5. Global Geomagnetic Perturbation Forecasting with Quantified Uncertainty using Deep Gaussian Process - Hongfan Chen (University of Michigan)