Welcome to the Edge Computing Group based in the Research & Development Center at the University of Puerto Rico at Mayagüez!
The Edge Computing Group (ECG) is directed by Dr. Wilfredo Lugo Beauchamp, Assistant Professor in the Department of Computer Science and Engineering at the University of Puerto Rico at Mayagüez (UPRM). The ECG focuses on developing tools to decrease resource requirements current computing methods consume before, during, and after deployment, including power, processing, and memory demands. We specialize in applying complex algorithms to run efficiently on resource-constrained environments, namely embedded systems such as IoT devices and wearable sensors. By optimizing computation at the edge, we aim to minimize data transfer, enhance privacy, and unlock new possibilities for real-time applications in areas like agriculture, healthcare, and environmental conservation.
Ongoing Research
Model Compression Engine for Wearable Devices on Skin Cancer
When it comes to saving a person's life, early detection of illnesses can make the difference between life and death. Machine learning models can be of great help in this case, but their complexity scales with their effectiveness, limiting the types of devices that can run said models. Our objective with this project is to develop a compact model that is capable of running in edge devices while maintaining an acceptable level of accuracy when it comes to detecting skin cancer. We seek to achieve this via a technique called transfer learning, where we take a reliable model that has shown to be effective in other areas, MobileNet V2 in our case, and we modify and train its outer layers so that it is capable of classifying pictures of a patient afflicted with skin cancer.
Addressing Memory Consumption in Edge Devices through Data Quantization
In Puerto Rico, frequent natural disasters disrupt healthcare systems, making it essential to develop diagnostic tools that work without relying on central networks. Our project focuses on optimizing deep learning models by reducing their memory and computational demands, allowing them to run efficiently on portable devices. We aim to do this through quantization, essentially converting 32-bit data to 8-bit, significantly cutting memory usage. This approach will be applied to a biomedical case of anemia detection through conjunctiva pallor analysis. By using lightweight models, we hope to create effective, low-power diagnostic tools that can function in areas with limited infrastructure, improving healthcare access during emergencies.
If you are a current or admitted student at UPRM and interested in joining our group, please email: wilfredo.lugo1@upr.edu
Our Supporters
