FURI | Spring 2026

Accelerating Convolutional Neural Network Inference with Posit Computing

Posit arithmetic offers a revolutionary alternative to traditional floating-point representations, delivering enhanced precision while improving memory efficiency across machine learning tasks. This project explores the practical implementation of posit-based systems in convolutional neural network inference, focusing on memory efficiency, processing speed, and model accuracy. By developing custom hardware-accelerated posit systems, this research will significantly reduce computational costs and enable faster inference on edge devices without sacrificing performance quality, democratizing access to state-of-the-art computer vision applications.