If you are interested in applying for summer 2022, please submit your application at this application form. For your reference and convenience, you can also check the PDF version of the form. Also, the deadline is only for application submission via the form. For recommendation letters, we understand they are not fully controllable by applicants, so it is fine if recommendation letters are sent a few days later than the application deadline.
The tentative time schedule of the eight-week summer 2022 program is 06/06/2022-07/29/2022.
Flyer for summer 2022 program. Please also check our Q&A webpage for application eligibility, requirements and additional information on the program. As required by the NSF REU program for all REU grants, only US citizen, nationals and permanent residents are eligible to apply.
Tentative Projects for Summer 2022
Team 1 Project: Big Data and Machine Learning Techniques for Atmospheric Remote Sensing
- Research Mentor: Dr. Jianwu Wang, Associate Professor of Data Science, Department of Information Systems
- Collaborator: Dr. Chenxi Wang, Research Scientist at Climate and Radiation Branch of the Laboratory for Atmospheres, NASA Goddard Space Flight Center (GSFC)
- RA Student: Xin Huang, PhD student, Department of Information Systems
Application background: Clouds have long been viewed as one of the most important components of the climate system because of their substantial impacts on the Earth’s radiation balance and global hydrologic cycle. Tremendous efforts have been made to reproduce cloud processes in models; however, the representation of ice cloud microphysical and macrophysical properties in current climate models remains one of the largest sources of uncertainty in predicting climate variability and changes. A prerequisite for cloud optical and microphysical property retrievals is identifying the presence of clouds, i.e., clear/cloudy classification. Additionally, characteristics such as cloud thermodynamic phase are needed as they can strongly impact the scattering and absorption properties of cloud droplets/particles. Until now, cloud detection and classification is still a difficult task since both horizontal/vertical structures and background (e.g., the surface and atmosphere) may strongly impact satellite observations. Research problem: A frontier research problem is to investigate how machine/deep learning algorithms could help cloud detection and classification from satellite data. Big data challenges: Many satellites generate very large volumes of observational data. For instance, NASA satellite Terra was launched in 1999 and has been in mission for over 20 years. Its MODIS instrument generates about 100 GB of level 2 Cloud Product data every day, which accumulates over 700 TB in total. Distributed and scalable machine/deep learning algorithms are needed to train over large scale datasets. Experiments and evaluation: We plan to measure prediction accuracy and execution scalability of designed machine/deep learning algorithms at our HPCF clusters, and compare their prediction accuracy with state-of-art physics based detection algorithms. Preliminary results: We have published preliminary results of this problem at  and . REU research project: In this project, we will extend our preliminary work to utilize larger volumes of training datasets by designing algorithms that are scalable on multiple computing nodes, especially GPU nodes and study how prediction accuracy could be further improved.
Team 2 Project: Big Data and Machine Learning Techniques for Medical Image Classification
- Research Mentor: Dr. Matthias Gobbert, Professor of Mathematics, Department of Mathematics and Statistics
- Collaborator: Dr. Jerimy Polf, Associate Professor in the Department of Radiation Oncology, University of Maryland School of Medicine
- RA Student: Carlos Barajas, PhD student, Department of Mathematics and Statistics
Application background: Proton beams’ primary advantage in cancer treatment as compared to other forms of radiation therapy, such as x-rays, is their finite range. The radiation delivered by the beam reaches it maximum, known as the Bragg peak, at the very end of the beam’s range. By exploiting the properties of the Bragg peak, it is possible to only irradiate cancerous tissues, avoiding any damage to the healthy surrounding tissues. A current major limitation of utilizing Bragg peak is the uncertainties in the beam’s position in the body relative to important organs that should not be irradiated. Research challenge: The Compton camera works by detecting prompt gamma rays emitted along the path of the beam. By analyzing how prompt gamma rays scatter through the camera, it is possible to reconstruct their origin. However, the raw data the Compton camera outputs does not explicitly record the sequential order of the interaction data which represents scatterings of a single prompt gamma ray, which makes reconstructions based on Compton camera data noisy and unusable for practical purposes. REU research project: This project will extend our preliminary work for the reconstruction of the Compton images for better deep learning model prediction accuracy. The team can adjust the size and number of residual blocks of the model, and the method of data concatenation which could have large impact on classification time and accuracy. Impacts to participants and the discipline: The participants will study effects of different deep neural network models and configurations on the accuracy of prompt gamma ray reconstruction and will learn about a cutting-edge medical treatment and the inherent need for improving its efficacy.