MSci student specialising in mechanistic interpretability of neural networks at Durham University. Founded Durham's AI Safety Initiative (DAISI).

Order by Scale: Relative-Magnitude Relational Composition in Attention-Only Transformers
Farrell, Theo, Patrick Leask, and Noura Al Moubayed
NeurIPS 2025 Workshop on Socially Responsible and Trustworthy Foundation Models

📊 Google Scholar Profile

Order by Scale

Implementation of research on relative-magnitude relational composition in attention-only transformers. Investigates how neural networks learn to order sequences by numerical scale, with analysis of learned feature representations and attention mechanisms.

Stance Classification

NLP project implementing transformer-based models for detecting argumentative stances in text. Uses fine-tuned BERT models with custom preprocessing and evaluation frameworks for multi-class stance detection.

Bipedal Walking with RL

Reinforcement learning implementation for training bipedal agents to walk. Uses PPO and A3C algorithms in PyTorch with custom reward functions and training visualizations.

• Cryptography Coursework - Implementation and analysis of cryptographic algorithms and protocols. Covers encryption techniques, security analysis, and practical applications.
• Natural Computing Algorithms - Exploration of bio-inspired and nature-based computational algorithms including genetic algorithms, neural networks, and swarm intelligence.
• Image Processing - Computer vision and image processing techniques using OpenCV and Python. Includes filtering, edge detection, and image transformations.
• Data Compression Coursework - Implementation of data compression algorithms including Huffman coding, LZW, and run-length encoding. Analyzes compression ratios and efficiency.
• Data Cleaning and Analysis - Data preprocessing, cleaning, and statistical analysis using pandas and Python. Demonstrates data wrangling and visualization techniques.
• Metaheuristics and TSP - Implementation and comparison of metaheuristic algorithms (genetic algorithms, simulated annealing, ant colony optimization) for the Traveling Salesman Problem.
• KNN vs Logistic Regression - Comparison of K-Nearest Neighbors and Logistic Regression classifiers on various datasets, with implementations from scratch and performance analysis.
• EPQ Calculator - My first major project, developed as part of my Extended Project Qualification (EPQ). A calculator application built to learn fundamental programming and UI design concepts.

• Ghost Marks in the Machine (DAISI) - Research project investigating unlearning techniques in language models. Implements methods for removing specific knowledge from trained models and evaluates their effectiveness.
• Durham AI Safety Initiative Website (DAISI) - Website for Durham's AI Safety Initiative, providing information about the organization, events, and research.
• Hands2Text (DurHack '23) - Hackathon project converting sign language to text using computer vision and machine learning. Built during DurHack 2023 hackathon.
• Context Eval - Evaluation framework for testing language models on context understanding and reasoning tasks.

• 💻  
• 🤖  
• 🛠️  
• 📱