Final Year Project

Project Title

Compression-based Perceiver Architecture

Description

Our project is research based. We propose a novel architecture for computer vision problems, building up from Google DeepMind's Perceiver, and our research question is to evaluate the performance of the Perceiver Architecture for classification when it is passed a latent representation of the image as an input. Our hypothesis is that the architecture will be independent of the input data-representation (adaptable to not just the images but also point clouds, audios, videos etc.) and will be many times more resource-optimized than its predecessor's architecture, the Perceiver, while still maintaining comparable accuracy metrics.