Optimization and Application of the Eyeriss Object Classification System
This project focuses on the optimization and application of a low power and offline image classification system developed by the Energy-efficient Multimedia Systems Group called Eyeriss. Eyeriss is a neural network accelerator that can be trained to complete object recognition tasks on streaming video at up to 35 frames per second. The first task of this SuperUROP will be to improve the system’s classification throughput with PCIE optimizations between Eyeriss and a NVIDIA TK1 development board. The second part of this SuperUROP aims to integrate the Eyeriss system onto a small RC car and to begin to explore the possible applications of this system in autonomous vehicles or other environments that could utilize fast and local image recognition.

I’m participating in SuperUROP because I feel that my project will give me experience in a cutting edge field. With vision processing becoming a widely used technology I’m glad my SuperUROP is giving me valuable experience in a field that may otherwise be hard to enter as an undergrad. My experiences on MIT Formula SAE and lab classes have prepared me well for this position by giving me practical hands on knowledge.