MIT EECS — Texas Instruments Undergraduate Research and Innovation Scholar
EyeSocket: A Hardware- Accelerated Convolutional Neural Network Module for Computer Vision
Deep learning-based computer vision promises drastically improved recognition accuracies when compared to conventional algorithms. Current implementations, however, consume significant amounts of power and energy. Our platform, the EyeSocket, implements convolutional neural networks (CNN) with a hardware accelerator that promises a substantial increase in efficiency, while maintaining nearly the same performance. EyeSocket is a small and integrated circuit board module, suitable for portable devices such as biped robots and quadcopters. Additionally, it is compatible with standard open-source computer vision tools such as Caffe, allowing for easy integration into existing projects.
I’m a course 6-2 junior passionate about building wearable devices and embedded systems. I’ve always been fascinated by the process of creating electronic systems, from PCB layout to bring-up. This project combines my interest in low power devices with machine learning, a topic I’ve always wanted to learn more about.