Thomas Desautels

Seeing the sights on a cold day in Greenwich

Welcome to my research webpage! During my graduate studies, I worked in the group of Joel Burdick at Caltech, in Pasadena California. I now work at the Gatsby Computational Neuroscience Unit of UCL under a scholarship provided by the Whitaker International Program. This webpage includes information on my research interests and publications as well as code from my work.

Research Interests:

My graduate work was on a project to use epidural electrostimulation to help patients with spinal cord injuries to regain the ability to stand. Our collaborators have published initial findings in a human patient in the Lancet, here. Since we use complex electrode arrays to deliver these stimuli, and these arrays have a great many parameters over which to optimize, we work with machine learning to rigorously search this space and provide an effective therapy along the way. My role focused on how to use Gaussian process-based active learning algorithms to select stimuli for therapeutic purposes.

In doing this work, I collaborated with Professor Andreas Krause of ETH Zurich, drawing upon his expertise in GP active learning. His website may be found here.


My PhD dissertation:

Thomas Desautels, "Spinal Cord Injury Therapy through Active Learning," PhD thesis, California Institute of Technology, 2014.

[From Caltech THESIS]

Our ICML 2012 paper:

Thomas Desautels, Andreas Krause, Joel Burdick, "Parallelizing Exploration-Exploitation Tradeoffs with Gaussian Process Bandit Optimization," In Proc. International Conference on Machine Learning (ICML), 2012.

[Published Version] [Long] [arXiv]

And its extension into a paper on a family of batch or delayed GP bandit algorithms:

Thomas Desautels, Andreas Krause, Joel Burdick, "Parallelizing Exploration-Exploitation Tradeoffs in Gaussian Process Bandit Optimization," Journal of Machine Learning Research, vol. 15, pp. 3873-3923, December 2014.

[Published Version] [Appendices] [At JMLR]

Our experimental paper on the use of GP bandit algorithms to control epidural spinal cord stimulation in rats:

Thomas A. Desautels, Jaehoon Choe, Parag Gad, Mandheerej S. Nandra, Roland R. Roy, Hui Zhong, Yu-Chong Tai, V. Reggie Edgerton, Joel W. Burdick, "An Active Learning Algorithm for Control of Epidural Electrostimulation," to appear in IEEE Transactions on Biomedical Engineering, 2015.



New version released, April 2014:

A MATLAB implementation of GP-BUCB, GP-AUCB, and other several other algorithms, as well as demonstration code associated with our ICML 2012 and JMLR 2014 papers is available upon request. I am working on web hosting to once again make it publicly available.