Search for: All records

Abstract Research-integrated instruction is widely used to place inquiry-oriented work inside required engineering courses. In practice, multi-course initiatives often rely on similar formats across departments, such as projects, teamwork, and iterative tasks, even though fields differ in how problems are framed and how evidence is evaluated. This paper reports descriptive results from a college-wide initiative at a historically Black university that embedded research-integrated activities into ten upper-level courses in Civil and Architectural Engineering (CAE), Mechanical Engineering (ME), and Electrical and Computer Engineering (ECE). Students completed a Course Elements survey in the first and final weeks of the semester and a Learning Gains survey at the end of the semester. Results show learning-gain composites above the scale midpoint in all three departments, with no statistically significant difference at the composite level. Clearer differences appeared in selected course elements tied to ownership and communication, including studentdesigned work, documentation, oral presentation, and data collection. Across courses, higher endof- semester engagement with research-integrated elements was associated with higher reported learning gains. Overall, the findings support program-level coordination around shared elements while leaving room for discipline-specific enactment. 

In-context learning is a remarkable capability of transformers, referring to their ability to adapt to specific tasks based on a short history or context. Previous research has found that task-specific information is locally encoded within models, though their emergence and functionality remain unclear due to opaque pre-training processes. In this work, we investigate the formation of task vectors in a controlled setting, using models trained from scratch on synthetic datasets. Our findings confirm that task vectors naturally emerge under certain conditions, but the tasks may be relatively weakly and/or non-locally encoded within the model. To promote strong task vectors encoded at a prescribed location within the model, we propose an auxiliary training mechanism based on a task vector prompting loss (TVP-loss). This method eliminates the need to search for task-correlated encodings within the trained model and demonstrably improves robustness and generalization.