SLMC //people: Adrian Haith//

SLMC homeSLMC researchSLMC peopleSLMC projectsSLMC newsSLMC coursesSLMC hiringSLMC contactIPABSchool of InformaticsUniversity of Edinburgh
Adrian Haith

Adrian Haith
I am a final year PhD student in the SLMC group at the School of Informatics, University of Edinburgh, supevised by Sethu Vijayakumar.

I am interesting in computational modelling of human motor control and adaptation. In particular, my PhD has focussed on how the motor system is able to adapt to different kinds of disturbances to performance - For instance, a kinematic disturbance like wearing prism goggles, versus a dynamic disturbance such as a force field applied to the hand.

Other research interests of mine include optimal control theory, machine learning and computational neuroscience.

My background is in mathematics (MA, Cambridge; Sidney Sussex College), machine learning (MSc, Edinburgh) and neuroinformatics (MSc, Edinburgh). I am funded by the UK EPSRC/MRC through the Neuroinformatics Doctoral Training Centre.

Adrian Haith

IPAB, Informatics Forum #1.43,
10 Chrichton Street,
Edinburgh EH8 9AB.
Tel: +44 131 650 8281

Research projects
Bayesian models of sensorimotor adaptation
Many kinds of disturbances can be responsible for a dip in motor performance. Distinguishing between these and identifying the true cause of an error is important to maintain performance. This problem can be framed as a Bayesian estimation problem - infer which disturbance caused the error given noisy observations and noisy motor execution. I have recently collaborated with Carl Jackson and Chris Miall at the PRISM Lab, University of Birmingham, to test whether human behaviour matches the Bayesian approach to solving this problem.

Cerebellar-based motor learning models
Various computational models have been proposed of the role of the cerebellum in motor control and adaptation. In particular, some models have proposed a feedforward connectivity in which the cerebellar output contributes directly to the motor command, while Porrill and Dean have recently advocated a recurrent architecture for cerebellar connectivity. There turn out to striking and complementary differences between the computational properties of these two models, with different architectures being best-suited to adaptation different kinds of sensorimotor disturbance.

Stochastic integrate-and-fire neural models
I have worked with Liam Paninski to develop efficient methods for computating the likelihood of parameters within a stochastic integrate-and-fire neuron model.

Publications, Talks and Posters