This project aims to address these financial and computational challenges by developing innovative performance measurement and analysis techniques tailored for deep learning workloads and encapsulating these techniques into a profiling toolkit (i.e., DLToolkit). Building upon existing open-source profilers, DLToolkit will offer scalable analysis, aggregation, and visualization of deep learning workloads, providing invaluable insights to scientists, thus fostering expedited innovation in scientific applications using deep learning.

Learning how to code is a key and challenging component in computer science (CS) education. Traditionally, the primary focus in programming courses has been on achieving functional correctness. However, there is a growing recognition of the importance of program performance, as evidenced by factors such as execution time, memory usage, and other metrics. This shift has garnered attention from both students and instructors, highlighting the need to incorporate performance considerations alongside functional correctness in CS education. This project will develop EduPerf, which aims to hoist program performance as the first-order metric in CS education via tightly integrating program performance analysis into different levels of CS courses for both students and instructors.

Many engineering students enrolled in STEM courses lack the skills needed to design novel methods for collecting data and managing its provenance to support scientific reproducibility and reliability. Without these essential skills, students struggle to validate prototypes and resort to expensive proprietary equipment with restrictive interfaces that may not actually collect the right type of data. While most curricula emphasize data analysis and visualization (two core tenets of data science), there is a paucity of emphasis on data collection, and to a lesser extent on data storage. Students are generally given representative data sets to analyze but are not given the proper tools to design systems which will acquire, collect and store physical measurements ��� critical experience necessary for success in later years, graduate school, and industry. To better connect engineering education with industrial practices and prepare our students for the STEM workforce of tomorrow, we propose a major change to the current pedagogy centered on open source/open standards and the Raspberry Pi. Adopting an open-source philosophy toward data collection will enable stability, security, interoperability, and reliability across a variety of engineering disciplines that acquire data, both in the classroom and in research labs.