Abstract
This paper summarizes the activities and outcomes of the AMALTHEA REU program, a collaborative summer research experience in Machine Learning hosted by the Florida Institute of Technology and the University of Central Florida. The program successfully integrated technical training with multi-tiered mentorship to produce significant academic contributions and bolster the STEM pipeline.
Introduction
Research Experience for Undergraduates (REU) programs are vital for strengthening the STEM pipeline, yet they face significant logistical hurdles, particularly when distributed across multiple campuses. The AMALTHEA program (Advances of MAchine Learning in THEory & Applications) was established to address the diversity gap in doctoral-level STEM fields and to develop best practices for collaborative research sites. By focusing on underrepresented minorities and women, the program aims to provide a cohesive community for geographically distributed faculty and students.
Methodology
The 10-week AMALTHEA program utilized a joint structure between the Florida Institute of Technology (FIT) and the University of Central Florida (UCF). Key components included:
- Machine Learning Primer (MLP): A 4-day intensive bootcamp that leveled the playing field for students from diverse academic backgrounds, covering core concepts like pattern recognition and regression.
- Collaborative Infrastructure: Weekly "All-Hands Meetings" (AHMs) featured technical talks and social "special projects" designed to build camaraderie across campuses.
- Tiered Mentorship: Research teams consisted of undergraduates, graduate student mentors, and faculty mentors, providing students with immediate technical support and exposure to professional research environments.
- Moodle Integration: The use of a course management system facilitated daily journaling and progress tracking across different sites.
Research Domains
Participants explored a wide range of high-impact Machine Learning domains, including:
- Algorithmic Improvements: Decision tree strategies and SVM training solvers.
- Computer Vision: Real-time gesture recognition and automatic license plate location.
- Speech & Audio: Wake-up word technology and interactive evolutionary music generation.
- Data Mining: Outlier detection and MapReduce-based grid systems.
Key Insights and Outcomes
A major factor in the program's success was the "stakeholder" effect, where mentors were deeply invested in research topics aligned with their own work. Additionally, the program demonstrated that social cohesion is a functional necessity for distributed teams.
Recruitment efforts successfully targeted underrepresented groups, ensuring that 50% of participants came from diverse backgrounds. This was achieved by leveraging a national network of affiliate universities.
Quantitative results from the first two years include:
- Participation: 20 undergraduate participants (7 in 2007, 13 in 2008).
- Academic Output: 5 conference papers and 1 journal paper co-authored by undergraduates.
- Pipeline Impact: Support for multiple Honors, Masters, and Doctoral theses.
- Student Sentiment: Participants unanimously recommended the experience, noting high levels of direct research engagement.
Conclusion
While the 10-week duration is short, the AMALTHEA model proves that collaborative REUs can be highly effective. Future work will focus on securing private sector funding and scaling the model to incorporate more institutions, testing the sustainability of this collaborative framework.
Summary updated