Dissertation Project - RULSTM Architecture

Grade Received: 1st (71%) 🥇

Overview

This project is designed around the improvement and developmenet of the RULSTM architecture. I have tweaked the original architecture to increase efficiency and also introduce newer architectural technologies from the field (AVT). In addition, I carried out a number of experiments to explore potential future research within the field of computer vision, as detailed within the adjacent research paper.

Features

• Tweaking sequence completion pretraining to improve performance on lower end machines.
• Longer time scale anticipation time experiments.
• Loss function experiments (MSE, Hinge, Kullback-Liebler and AVT) to increase efficiency.
• Optimiser function experiments.
• Observation Time optimisations from original architecture.
• Integrating backbone technology from AVT architecture.

Important Links

Here are some key links to explore:

• Original RULSTM Repository
• Original RULSTM Paper
• My RULSTM Research Paper

Authors

• @oliciep