I am a Visiting Researcher at the Alan Turing Institute in London. Previously I was a Research Fellow at UCL Computer Science. Before that I obtained my DPhil in Computational Statistics and Machine Learning at the University of Oxford.
My work is on combining deep learning with Bayesian statistics. The resulting class of models, deep generative models, enable us to scale Bayesian approaches to large datasets and complex data like images.
I am interested in three broad areas:
-
The robustness of these models, to missing data and to adversarial attack.
-
Building highly structured models that enable us to learn representations that are in some sense simple, even human-interpretable.
-
How these models act when scaled up to have deep hierarchies of learnt representations.
Publications
-
Variational Autoencoders: A Harmonic Perspective, 2022, AISTATS, A Camuto, M Willetts
-
Certifiably Robust Variational Autoencoders, 2022, AISTATS, B Barrett, A Camuto, M Willetts, T Rainforth
-
Multi-Facet Clustering Variational Autoencoders, 2021, NeurIPS, F Falck*, H Zhang*, M Willetts, G Nicholson, C Yau, C Holmes
-
Relaxed-Responsibility Hierarchical Discrete VAEs, 2021, NeurIPS Bayesian Deep Learning Workshop, M Willetts, X Miscouridou, S Roberts, C Holmes
-
Improving VAEs’ Robustness to Adversarial Attack, 2021, ICLR, M Willetts*, A Camuto*, T Rainforth, S Roberts, C Holmes
-
Learning Bijective Feature Maps for Linear ICA, 2021, AISTATS, A Camuto*, M Willetts*, B Paige, C Holmes, S Roberts
-
Towards a Theoretical Understanding of the Robustness of Variational Autoencoders, 2021, AISTATS, A Camuto, M Willetts, S Roberts, C Holmes, T Rainforth
-
Semi-Unsupervised Learning: Clustering and Classifying using Ultra-Sparse Labels, 2020, IEEE Conference on Big Data – Special Session on Machine Learning for Big Data, M Willetts, S Roberts, C Holmes
-
Explicit Regularisation in Gaussian Noise Injections, 2020, NeurIPS , A Camuto, M Willetts, U Şimşekli, S Roberts, C Holmes
-
Disentangling to Cluster: Gaussian Mixture Variational Ladder Autoencoders, 2019, NeurIPS Bayesian Deep Learning Workshop, M Willetts, S Roberts, C Holmes
-
Semi-Unsupervised Learning using Deep Generative Models, 2018, NeurIPS Bayesian Deep Learning Workshop, M Willetts, S Roberts, C Holmes
-
Statistical machine learning of sleep and physical activity phenotypes from sensor data in 96,220 UK Biobank participants, 2018, Nature Scientific Reports, M Willetts, S Hollowell, L Aslett, C Holmes & A Doherty
* equal contributions
Working Papers
- On Algorithmic Stability in Unsupervised Representation Learning, 2021, arXiv preprint, M Willetts, B Paige