Rishab Nayak is a student at Boston University, studying biochemistry and biomedical engineering with a passion for the application of computer vision and modern technology in medical diagnostics. His goal to promote the implementation of computer vision in medical diagnostics has led him to foray into the field, designing image recognition algorithms and pattern detectors. By joining the Wolfram Summer School, he hopes to learn how to better design these algorithms and better his understanding of the field. His aim is to hopefully help build the future of diagnostics by miniaturizing diagnostic machinery and making its adoption cost-effective. He is interested collaborating with others and looking to engage in multifaceted projects.
Project: A Performance Analysis of Multiple Neural Networks to Identify Plugs and Connectors from an Image
Analyze the network performance of different neural network approaches to solve the same problem: identify plugs and connectors from its image. This would allow us to understand the implications of using different network frameworks on accuracy and compute time and to understand the tradeoffs.
Main Results in Detail
On analyzing the data, it was found that the Ademxapp v2 network performed the best, with an accuracy of 91%, but also was one of the slowest networks, having an evaluation time of 1.10 seconds. The fastest network was the ImageIdentify v2 network, with an evaluation time of 0.062 seconds; however, it compromised on accuracy, dropping to 82%. From the data collected, it appears that there exists a direct relationship between accuracy and speed for neural networks. As the depth of the network increases, the accuracy increases, resulting in an increase in evaluation time.