|

MSc Cognitive and Decision Sciences Graduate & Deep Learning Practitioner

Russell de Roeper

Summary.

I am a recent Computer Science and AI graduate from Loughborough University and a former software developer at Oscore Ltd. I have always been captivated by the idea of thinking machines and the human mind. These interests have led me to become a MSc Cognitive and Decision Science student at UCL, where I explored the human mind from a computational perspective. I was awarded a Distinction with a Dean's List Nomination for my MSc studies.

I am looking to apply my studies in the Cognitive Sciences to tackle continual learning in deep neural networks, particularly in Large Language Models. Alongside my MSc degree, I am working part-time as a freelance software developer and piano tutor. In the future, I want to continue exploring my research interests, and apply it to real-world problems.

Download CV

Coding Projects.

Robot with Zed Camera

Point Cloud Optimiser

Developed a genetic algorithm to search the space of possible settings for the Zed 2 Stereo Camera to optimize point clouds at the Intelligent Automation Centre.

Genetic Algorithms 3D Vision Python
View Presentation
VAE Latent Vector Reconstruction

Variational Autoencoders

Recreation of the 2018 World Model's VAE architecture applied to the Cifar10 dataset using PyTorch. Explored the effects of varying latent space dimensionality.

PyTorch Generative AI Deep Learning
Learn about World Models
Bullet Dodger Game

2-Player Bullet Dodger

A fast-paced 2-player arcade game built with PyGame. Features custom projectile physics, originally purposed as a Reinforcement Learning environment.

Python PyGame Physics
Art Recommender

Artwork Gallery Recommender

A client wanted a new website feature to help explore their artwork collection. I built a proof-of-concept artwork recommender using a pretrained CNN at Oscore Ltd. This feature recommends similar artworks from a gallery based on an input image by computing vector distances.

CNN ImageNet Recommender

ChatGPT Ticket Helper

Internal tool built for Oscore Ltd. Uses an Azure DevOps webhook and a Flask API to automatically generate ChatGPT responses with potential solutions for new work tickets.

Azure DevOps Flask API OpenAI API
LTN + DDQN Architecture

Neuro-Symbolic Reinforcement Learning

BSc Final Year Project exploring how to build an interpretable RL agent within Minihack using Logic Tensor Networks (LTN). Combines formal logic embedding with Deep Q-Networks.

RL Neuro-Symbolic AI LTN
Read Paper

Let's Connect.

I am currently looking to apply my studies to real-world problems. Feel free to reach out to me for opportunities or just to say hi!