Mobilizing the Emerging Diverse AI Talent (MEDAL) through Design and Automated Control of Autonomous Scientific Laboratories
Dr. Sumit Kumar Jha, University of Texas at San Antonio
Dr. Arvind Ramanathan, Argonne National Laboratory
Dr. Sreenivasan Ramasamy Ramamurthy, Bowie State University
Dr. Sunny Raj, Oakland University
Dr. Sathish Kumar, Cleveland State University
Dr. Giri Narasimhan, Florida International University
Dr. Rickard Ewetz, University of Central Florida
The Mobilizing the Emerging Diverse AI Talent (MEDAL) project is a collaboration between University of Texas at San Antonio, the Florida International University, the Argonne National Laboratory, Bowie State University, Oakland University, Cleveland State University, and the University of Central Florida. The University of Texas at San Antonio, Florida International University, and the University of Central Florida are Hispanic-Serving Institutions (HSIs), and Bowie State University is one of the Historically Black Colleges and Universities (HBCUs). Cleveland State University and Oakland University are doctoral universities with high research activity.
The MEDAL project aims to train faculty, doctoral, graduate, and undergraduate students at the HBCU, three HSIs, and two urban universities with high research activity on contemporary artificial intelligence (AI) research relevant to the Department of Energy, including topics such as transformers, large language models, pre-trained visual and scientific models, and data-driven control of autonomous scientific labs. The project comprises four distinct tasks that aim to enhance learning outcomes on these topics by catering to diverse learning preferences. These tasks include (i) the adaptive delivery of video lectures using AI, (ii) the automated evaluation and related feedback generation assisted by large language models, (iii) the use of deep reinforcement learning and pre-trained models for autonomy in scientific labs, and (iv) the sustained dissemination of our efforts through outreach through summer workshops and asynchronous online AI classes. By implementing these tasks, the project aims to train students to create a new diverse workforce capable of contributing to AI engineering and research in line with the technical requirements of the Department of Energy (DOE).
The project team seeks to pursue scientific and technical innovations to support its educational and outreach plans. The project team, including Ph.D. students and undergraduate researchers, is studying topics of contemporary research, such as (i) deep reinforcement learning for control of autonomy in scientific labs, (ii) automated verification of image responses from codes in AI classes, (iii) algorithmic generation of explanations for incorrect images produced by AI programs, (iii) designing a calibration metric for automatically generated feedback on AI-related code fragments, (iv) enhancing privacy and fairness of the underlying AI models deployed in our education activities, (v) fine-tuning and pruning pre-trained visual models for robotic control in autonomous scientific lab cyber-physical system test beds, and (vi) robustness metrics for pre-trained models used in our automated evaluation and feedback generation process. The research team, comprised of a collaborative group of six academic institutions and the Argonne National Laboratory, is actively investigating these fundamental research problems.
The MEDAL project aspires to provide AI education and training opportunities to diverse and historically underserved populations and promote DOE-relevant research by leveraging online learning and recent advances in large language models and other large pre-trained models for visual and scientific information.
