Model-based Signal Processing and Sparse Priors

Description:

In my Master’s thesis at ETH Zurich, I used statistical signal processing methods for separating positional eye movement measurements into different types (saccades, smooth pursuit, and fixation eye movements). I developed a novel approach to precessing eye movement signals based on estimating signals in a mechanistic physiological model of the eye muscles. Apart from signal separation, the framework is also able to estimate the neural inputs into the eye muscles from the positional measurements.

Methods:

• Factor graphs: They are a powerful probabilistic framework for working with structured models and has many applications, e.g., state space models, image models, error correcting codes, optimal control.

• Sparse Bayesian learning: This is a widely applicable and efficient method for modeling and estimating sparse (non-gaussian) priors in a probabilistic framework, which we applied to factor graphs.f

Specials:

I made especially two contributions in my thesis: Firstly, the usage of a new sparsity prior within the factor graph framework, which allowed to appropriately set the sparsity level. Secondly, the usage of existing mechanistic eye movement models, which have been developed since the 1980s, for solving this problem.

Technology:

Matlab

Links:

• Journal paper: Model-based separation, detection and classification of eye movements

• Code github