Improving AI Performance at SURP

Improving AI Performance at SURP

Improving AI Performance at SURP

The Summer Undergraduate Research Program (SURP) provided Rowan CS students with an opportunity to apply the skills they’ve learned by working on a research project. One such student was Tyler Casas, who presented the findings of his research project at the SURP Poster Session in July. For his project, he developed ideas for improving AI performance on Internet of Things devices.

Now, it’s no secret that technology is constantly evolving. Even the slightest advancement in technology can make waves in people’s way of life. Take cellphones, for example. They didn’t even exist a century ago, but nowadays it’s hard to imagine life without them. That said, even today’s awe-inspiring tech has its limits. For example, small devices tend to have very limited computational power compared to their larger counterparts. So, how can we ensure that small devices keep up with the constant forward march of progress? Tyler Casas seems to have found a potential solution.

Using an Internet of Things infrastructure, it’s possible to offset the computational burden that small devices have to carry, by wirelessly connecting them to more powerful devices and performing the computations there instead. By outsourcing heavy computations in this fashion, it would be a lot easier for smaller devices to keep pace with larger devices and perform computationally intense operations with more efficiency.

In order to test out this idea, Tyler got to work implementing a variety of wireless, server-based configurations in order to compare their computational capabilities to that of a PI-only local machine. A few different AI models were used for this testing process, including AlexNet, MobileNetV3, SqueezeNet, and VGG.

Just as Tyler expected, the performance of the PI-only local machine paled in comparison to the high-speed efficiency of the wireless, server-based configurations. The server-based models were astronomically faster, with AlexNet being the fastest. These results not only act as a great showcase of the merit of server-based computation, but also go to show just how big of a difference it can make compared to a local machine handling the task all on its own.

Even with all this progress so far, this is only the beginning. Tyler predicts that this infrastructure for AI-based computation could have a vast array of potential applications, from things like algae build-up detection to better facial recognition software. The possibilities are endless!

Tyler’s research project already shows a lot of promise, as do the other students’ projects here in the Computer Science department. Be sure to keep an eye out for future articles as we continue to showcase the incredible accomplishments of Rowan’s CS students. Whether it’s SURP or the CS Department at large, there’s no shortage of interesting projects worth examining.

Written by Cole Goetz  |  Posted 2022.10.17