Once again, the Provost's Office has provided funding through the Learning Analytics Fellowship program to faculty and staff for the 2025-26 academic year. Details of the new fellows’ proposals are provided below:

Sarah Bass(Chemistry) will reprise (and dive deeper into) a Spring 25 DoIT & FDC panel presentation on her use of NotebookLM to develop "Virtual Prof. Bass," an AI-driven tool that guides students in her CHEM 101 course to proactively engage with course materials. Specifically, she will focus on at-risk students she studied further this summer in Carnegie Mellon’s LearnLab: Compared to students who take the CHEM 101 & 102 sequence in Fall and Spring, “off-track” students who do the opposite tend to underperform on standardized American Chemical Society (ACS) concept inventory exam questions embedded in her unit, midterm, and final exams. In addition to her 24/7 “Virtual Prof Bass,” she will market to and support “Off-track” students by offering focused practice and success strategies during the CHEM 101 common exam time on Friday afternoons from 4-6 p.m. Note: In addition to her LA CoP renewal, Bass’ research and practice are supported by a second year of funding from Digital Promise ChemCore, to support faculty use of the REAL Chem courseware developed by Arizona State University and Carnegie Mellon University. 

Note: Bass will present her work to date in a talk, "Using AI and Analytics to Assess and Supplement Students' Prior Knowledge Needed to Succeed in Large STEM Courses," on Tuesday, October 14, at Noon, co-sponsored by DoIT and the Faculty Development Center (FDC). This will also serve as the first Learning Analytics (La) Community of Practice (CoP) talk for the Fall 2025 semester.

Neha Raikar(Chemical, Biochemical, and Environmental Engineering) will test whether anonymous grading can reduce bias and student anxiety. The study asks three things: (1) Does hiding names lower real or perceived bias? (2) Does the effect change by assignment type—auto-graded vs. instructor-graded, multiple-choice vs. free-response? (3) Does anonymity help students take risks, stay engaged, and continue in the course? Assignments will be randomly split into two groups: one graded with names, one without. Graders will use the same scoring guides and check that their scores match, and the team will compare results with basic statistics. Short interviews will gauge the time needed, benefits, costs, and any side effects. Students will take brief surveys before and after assessments about fairness and why they feel that way. Reports will share group trends only, not names. The expected result is fairer grading, higher student confidence, and clearer insight into what shapes scores.

Liang Zhu (Mechanical Engineering) is using developed conceptual questions at the start and end of an undergraduate heat transfer course to see where students begin strong, where they struggle, and how their understanding evolves across key course concepts. Keeping the same instructor ensures a consistent view, and early results show noticeable gains. The team will study patterns in the answers to spot common stumbling blocks, judge whether any questions are unclear or too easy, and confirm that the set fairly represents the course. Those insights will drive concrete improvements: clarifying or replacing weak questions, adding timely examples and practice where misconceptions persist, and fine-tuning pacing and emphasis. The question set will be refreshed into a balanced collection that better distinguishes levels of understanding and yields clearer feedback for teaching. The revised questions will be tried in a future offering of the course and accompanied by a brief student survey to gather perceptions and suggestions. Results will be shared with the funding office and at local education venues, align with departmental goals for cost-effective, evidence-based teaching, and inform similar updates in related courses with potential to scale more broadly.

These one-year learning analytics awards are renewable pending receipt of a final report, paper submitted for publication or conference presentation. In addition to use of UMBC’s Report Exchange (REX) data warehouse, and a Tableau “viewer” license, faculty recipients can consult with staff from Analytics and Business Intelligence, Instructional Technology and Institutional Research and Decision support (IRADS). 

For more information about this year’s workshops, speakers and another learning analytics mini-grant call for proposals to be announced in Spring 26, please visit doit.umbc.edu/analytics/community. 

By John Fritz & Tom Penniston

Learning What Matters: Causal Time Series Modeling for Arctic Sea Ice Prediction

Authored by Emam Hossain (PhD Candidate), Dr. Md Osman Gani

Abstract:

Conventional machine learning and deep learning models typically rely on correlation-based learning, which often fails to distinguish genuine causal relationships from spurious associations, limiting their robustness, interpretability, and ability to generalize. To overcome these limitations, we introduce a causality-aware deep learning framework that integrates Multivariate Granger Causality (MVGC) and PCMCI+ for causal feature selection within a hybrid neural architecture. Leveraging 43 years (1979–2021) of Arctic Sea Ice Extent (SIE) data and associated ocean-atmospheric variables at daily and monthly resolutions, the proposed method identifies causally influential predictors, prioritizes direct causes of SIE dynamics, reduces unnecessary features, and enhances computational efficiency. Experimental results show that incorporating causal inputs leads to improved prediction accuracy and interpretability across varying lead times. While demonstrated on Arctic SIE forecasting, the framework is broadly applicable to other dynamic, high-dimensional domains, offering a scalable approach that advances both the theoretical foundations and practical performance of causality-informed predictive modeling.

Check the full paper here: https://www.alphaxiv.org/abs/2509.09128

Join us in congratulating the authors on their hard work and getting their paper accepted by the IJCAI 2025 Workshop on AI for Time Series Analysis (AI4TS)