I am a PhD student in the Machine Learning Group, supervised by Carl Rasmussen. I am funded by an EPSRC Doctoral Training Award. Before starting my PhD, I completed an MEng in Engineering from Jesus College, Cambridge.
I am broadly interested in probabilistic inference and machine learning. I have taken a liking to variational inference in particular, and my aim is to to extend existing methods to allow for a more flexible trade-off between accuracy and time complexity. I am paricularly interested in applying these methods to the learning of dynamical systems in order to provide alternatives to asymptotically correct particle MCMC methods and current deterministic approximations based on simple Gaussian approximations.
I have worked on projects on scaling up variational inference in GPs, pseudo-marginal MCMC for exact Bayesian inference of GP hyperparameters and latent variable models designed to highlight new aspects of data (see below).
Variational Inference for Latent Variable Modelling of Correlation Structure
We introduce the Wishart Process Latent Variable Model. We aim to present a latent variable model which captures a different aspect of the data than existing methods by modelling the covariance between dimensions of observations, rather than the value of an observation. Additionally, this also allows us to compare whole datasets based on their covariance structure.
Mark van der Wilk, Andrew G. Wilson, Carl. E. Rasmussen
Workshop on Advances in Variational Inference, NIPS, 2014
[pdf, arXiv, Software, BibTeX] - To be released
Distributed Variational Inference in Sparse Gaussian Process Regression and Latent Variable Models
Here, we scale up variational inference in sparse GP regression and GPLVMs by distributing the calculations among independent computers. We run on large datasets and show that this benefits performance.
Yarin Gal*, Mark van der Wilk*, Carl E. Rasmussen
[pdf, arXiv, Software, BibTeX]
Workshop on New Learning Models and Frameworks for Big Data, ICML, 2014
[arXiv, Software, BibTeX]
Variational Inference in the Gaussian Process Latent Variable Model and Sparse GP Regression -- a Gentle Tutorial
An in depth tutorial on the detailed derivations of the variational inference scheme for the GPLVM and sparse GPs. We attempted to collect various insights from across the literature into one easy reference document.
Yarin Gal, Mark van der Wilk
* Joint first author.
I have supervised a variety of courses for the Cambridge Engineering course, including Linear Systems & Control (2nd year), Probability (2nd year) and Medical Imaging and Computer Graphics (3rd year).
If you are looking for notes I have promised, take a look here.