Adaptive Anchored Inversion
Adaptive Anchored Inversion for Gaussian random fields using nonlinear data is my final (and representative) academic publication.
It is a computation-intensive statistical method tackling a problem that is a longtime fundamental challenge in several disciplines. I conceived the approach and developed it over four years’ of intense work, undertaking three or four overhauls to the methodology, each time lifting it to a new level. (Academics do not use the term “invent”, but this is just that.) I developed several software packages along the way to implement the method. Results of example applications are very compelling.
Some strategies in this work are very familiar and important to Machine Learning, including:
- Bayesian statistics
- iterative algorithm
- kernel density estimation (I have a separate publication on this as a by-product of the main-line of research)
- dimension reduction
- automatic adaptation
- loss function
This is a piece of research work that I am proud of. I believe it continues to be a meaningful contribution to the field of its subject matter. (However, it has not gained much traction, because I left academia soon after publishing the work.) I try to make the methodology useful, or at least visible, to others via a web service. The idea is that the modeling code runs behind a web service, whereas the user interacts with it through web UI or API calls. The web site has some management and demonstration capabilities, including dynamic plots. This idea is functional on a proof-of-concept level but not polished. Progress is slow due to lack of time.
Example client code resides in the Github organization “anchored-inversion”.
Open-source Coding Projects
I do not have any large software project in the open. I have a github account, which contains some small experiments and utilities.