Will Leeney

I’m passionate about AI, Open-Source Software, Crypto, Crypto, Yoga and Basketball.

AI Generalist I completed my PhD in Artificial Intelligence at the University of Bristol supervised by Dr. Ryan McConville. My thesis examined current evaluation practice for machine learning, exploring how evaluations can misrepresent algorithms performance. I have spent years rigorously (I know LLMs love this word, but it is the most appropriate) comparing models against one another, so I know the importance of proper benchmarking practice (and subsequently I know how to get the best out of each model). These learnings apply in all machine learning problems because without a consistent evaluation you cannot know which algorithm you should employ. If you have any questions or help designing frameworks, then please reach out via email at will dot leeney at gmail dot com or on LinkedIn. If you want to take a look at this in the form of a CV then click here.

Supervision Throughout my academic journey, I have actively engaged in mentorship, guiding over 50 machine learning projects for final year undergraduates. I helped students formulate research problems, design experiments, and effectively communicate their findings through written reports and presentations.

Thesis Unsupervised Graph Neural Networks: Training Strategies and Evaluation Principles. This will be made available for public consumption shortly.

  • I demonstrated the need of a framework for more reliable evaluations showing the importance of experimental conditions.
  • I proposed empirically quantifying the sensitivity of performance comparisons to randomness.
  • I showed that it is possible to train GNNs with modularity to bypass a supervised optimisation in model selection or hyperparameter tuning.
  • I extended the framework to a new scenario for community detection, showing that multiple clients can collaborate on learning.

Areas of Study

  • Machine Learning Evaluations
  • Unsupervised Learning
  • GNNs
  • The Role of Randomness
  • Federated Learning
  • Community Detection

I’m interested in AI as I believe we can improve the lives of all humans through practical machine learning. It’s also pretty fun seeing sci-fi dreams come to life. The applications of my research can be to improve latent representations for enhancing search; to encode patient information for health care to predict drug treatment recommendation; to help detect fake news on social media networks; to find financial indicator signals for price prediction trading algorithms.

Software Tools

  • python
  • julia
  • typescript
  • latex
  • git
  • shell scripts
  • C
  • java
  • html
  • css