ADSA Annual Meeting - 2024

This session will introduce the Virginia model of data science, also known as the 4+1 model, a broad framework designed to shape conversions about teaching and doing research in the field. We will review the history of predecessor definitions and models, such as Conway's Venn diagram and CRISP DM, as well as Donoho's more controversial model of "greater data science." We will then demonstrate the derivation of an invariant structure from a persistent theme in these models--the concept of a data science pipeline as a way of organizing thinking around the field. Among the topics covered will be the ongoing tensions in the field, exhibited by differences between earlier models, the subfield of data design, and the influence of the model on how data science is taught at UVA. Of particular interest to us is design, which relates to both visualization and data as a form of representation. We hope to engage a lively discussion.

In the evolving landscape of our data-driven society, integrating ethical considerations with practical problem-solving in data workflows is crucial in both academic and industry contexts. This session will share our ongoing curriculum enhancement project at Indiana University Indianapolis, which incorporates a data ethics and data justice framework into the pedagogical strategies and teaching practices of undergraduate Data Science (DS) and master's level Library and Information Science (MLIS) curricula. Utilizing the Data Science Ethos Lifecycle, our goals are to: 1) update Program Learning Outcomes (PLOs) for DS to emphasize the human and sociotechnical dimensions of data work; and 2) create sample assignments, to be piloted in Spring 2025, that synergize ethics with technical content. We will present our updated PLOs and sample assignments for courses such as Data Policy and Governance, Social Media Data, and Public Library Management, and discuss: 1) using Data Science Ethos Lifecycle to develop a data science education curriculum that integrates ethical considerations with technical skills, transitioning from traditional compliance-focused education to a practical, problem-solving approach; 2) encouraging and assisting instructors from diverse backgrounds to incorporate a focus on critical thinking and ethical reasoning into their courses and assignments; and 3) enhancing career opportunities for students, particularly those from underrepresented groups and those with lower tech literacy, by offering enriched learning experiences. This session will highlight the role of human-centric education in preparing future data professionals who are not only skilled in technical aspects but also adept at addressing ethical challenges in the field.

In this session, we will explore challenges in transitioning AI advancements from research to practical social impact applications. We will share insights and case studies from our work in public health and conservation, highlighting successful strategies to overcome barriers, with a special focus on gaps in AI supporting high-stakes decisions. The session will begin with a brief presentation framing the topic's importance, then discuss advances in machine learning, multi-agent systems, uncertainty quantification, and data quality assurance, and how these have been successfully deployed to support the Centers for Disease Control and Prevention (CDC) and conservation NGOs. This will be followed by an interactive discussion with the audience aiming to foster collaboration, share experiences, and generate practical solutions.

Grounding data science in societal impact and real-world problems is core to our mission at the University of Chicago Data Science Institute. We integrate community-centered data science across research and educational programs, with the goal of creating student education and career pathways in social impact data science, while meeting the data and technical needs of our organizational partners working at the frontlines of change. For example, in the Data Science Clinic, students work with real-world clients, including social impact organizations, debating data science ethics and developing critical thinking alongside technical skills. Many collaborations emerge from the 11th Hour Project, for which the DSI serves as a centralized hub for nonprofit grantees in four main impact areas: energy, food and agriculture, human rights, and marine technology. We build long-running partnerships with these organizations that can span across multiple years. Recent examples include the interactive data tool PalmWatch, developed in collaboration with Inclusive Development International, which examines the environmental harm caused by palm oil production in countries around the world; and the Grocery Gap Atlas, in partnership with RAFI-USA, a tool that visualizes inequities in food access and corporate concentration in the grocery market at the census tract-level nationwide. Highlighting these projects as case studies, we will discuss why partnerships with social impact organizations are important to data science researchers and our communities, and what makes a successful data science project partnership.

The rapid rise of AI related jobs in the government and private sectors requires a careful classification of the knowledge, skills, and ability (KSA) that candidates need for filling such jobs and a description of suitable academic training. The ADSA community will collaborate with employers to first develop a taxonomy of learning objectives and competencies developed by AI and data science related educational efforts. We will concurrently develop a mapping of these to the KSA required by prospective employers.

Desired Outcomes:

I. From the training side (Taxonomy):
A taxonomy of topics and student learning objectives (SLO) of programs in data science and AI at different levels – certificate, bachelors, graduate certificates, master’s, and PhD.

II. From the employer Side (KSA):
A mapping between knowledge and skills for different data science and AI job categories to the created taxonomy. Establish competency levels and mechanisms for evaluation in the described knowledge and skills underlining the importance of education on “responsible data/AI citizens” in the face of algorithm bias and security issues around data.

The proliferation of AI technology in the domain of policing continues to increase despite public criticism from advocacy groups, local communities, and activist organisations. This session invites concerned researchers and practitioners to discuss possibile avenues to conduct algorithmic impact asessments of the use of AI algorithms in the policing domain. How can we wholistically approach the continuing developements such as in the use of facial recognition, automated license plate readers, biometric data collection, acoustic gunshot detection, predictive policing, LLM-generated police reports and so on? What are the risks and limits invovled in choosing to or not to work with police in order to conduct such audits? What kinds of data should be used, and how should we be critical of their provenance and bias while attempting to make use of them? What analytic frameworks e.g. causal analysis, statistical methods, or other forms of measurement should be used?


To scaffold the discussion, we will begin with a brief presentation on some of the ongoing work conducted at the Blue Data Lab on related problems and questions. The discussion will focus on developing new methodological approaches to the above, potential collaborations across discplines, and generating new questions and ideas.

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.

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)