Marlow Eliot

A boids flock from my flocking project. Move your cursor through it.

Marlow Eliot

MSc student in AI for Sustainable Development at UCL. I'm interested in applied machine learning and research, building systems that find structure in messy real-world data, and in helping governments regulate AI safely and keep pace with it.

Education

Now · 2026 – Sep 2027

University College London

MSc Artificial Intelligence for Sustainable Development

Modules include Accountable, Transparent and Responsible AI; Deep Representations and Learning; and Probabilistic Modelling.

2023–26

University of Sussex First-Class

BSc (Hons) Computer Science and Artificial Intelligence

Modules included Applied Machine Learning, Neural Networks and Advanced Natural Language Engineering.

Selected to represent the university in an international COIL debate on the societal impacts of AI.

Projects

2025–26

Self-supervised behaviour discovery Dissertation · 88/100

Python · PyTorch · HDBSCAN · UMAP · scikit-learn

An end-to-end pipeline that finds behavioural states in 7.4 million frames of 3D movement data from 52 tracked bumblebees, with no manual labelling. A convolutional autoencoder learns compact representations of movement, and HDBSCAN clusters them into behaviours.

A systematic architecture search cut reconstruction loss to about 1.5% of the baseline. It also showed that better reconstruction does not guarantee more stable clusters.

2025

Evolving flocking behaviour

Python · PyBullet · evolutionary algorithms · NumPy

Rebuilt Reynolds' boids flocking model in a PyBullet physics simulation, with randomly placed obstacles so evolved behaviour could not overfit to one layout. Benchmarked four evolutionary algorithms (hill climber, microbial GA, spatial GA and island-model GA) on tuning the flock's parameters.

Proposed a stratified model that gives agents on the edge of the flock different rules from those inside it, challenging Reynolds' assumption that every agent behaves identically. It outperformed the standard model under all four algorithms.

Experience

Jan–Apr 2026

World Vision International

Assisted a technical advisor for the Middle East and West Asia region

Built an environmental degradation database designed to be handed to AI agents. It links each environmental issue to its evidence: tracking data, surveys, studies and papers. We tested it on Armenia, to help field teams identify local issues and choose responses.

2024–26

The Badger, Sussex student newspaper

Senior Print Editor & Treasurer; previously Science & Technology Editor

Ran editing and production for a monthly paper with a 1,500-copy print run and managed its budget. The Science & Technology section was nominated for Best Science & Technology Publication at the 2026 SPA Awards.

2022

RSPCA, Huddersfield branch

Web design work experience

Reviewed the branch website and proposed a redesign of its adoption pages.

Selected writing

A selection of pieces from The Badger and Substack.

2026–

Substack

The Pasturage Institute

Co-founded with Sonny Wilcock

Non-partisan essays on politics, philosophy, and policy. The first, Apolitical Politics, sets out our approach: looking at how power shapes political decisions.

Sep 2024

Marlow's Substack

Examining the information processing model

An essay testing how well the computer-like model of the mind explains perception, memory and creativity, and how it compares with embodied, situated accounts of cognition.

Sep 2024

The Badger · Science & Tech

Running out of Data?

Why language models may exhaust usable human-written text, what curriculum learning and synthetic data can and can't fix, and the privacy questions that scarcity raises.

Skills

Machine learning
PyTorch, scikit-learn, self-supervised and unsupervised learning, clustering (HDBSCAN, UMAP), evolutionary algorithms
Programming
Python, NumPy, pandas, SQL, JavaScript, HTML/CSS, Git, PyBullet
Other
Writing and editing, budgeting, public debate