Researching new graph-based latent embedding methods for learning community level probabilistic graph ensembles from partially observed, high-dimensional bipartite graphs.

​

Results will be applied to predicting mitochondrial mutant pathogenicity from single-cell sequence data and scoring patient genetic health, providing a finer grained disease risk metric than existing bulk tissue methods.

Developing pySPoC, a Python package for large-scale, configurable and extendable automated feature extraction of point cloud datasets.

​

Use cases include:

• Dataset classification.

• Model selection.

• Model pre-training.

• Manifold learning.

• Self-similarity detection.

To what extent do datasets of all shapes, sizes and sources concentrate around manifolds of intrinsically lower dimensionality? And how much can one infer from the properties of this hypothetical manifold?

​

While the Manifold Hypothesis is provably false, researching the degree to which data adhere to well behaved generating processes offers insight into why methods such as data compression and dimensionality reduction are so effective in practice.

Development of new signal extraction method offering improved power spectrum recovery and robust to heteroskedastic noise in haptic EEG data.

​

With the method fully implemented in Python, work is now focused on documenting its performance, write up and publication.