Statistical Machine Learning

New experiments frequently mean entirely new statistical data reduction strategies are needed. We develop these to analyze our single-molecule and microscopy image data, frequently drawing on concepts from statistics and information theory.

Photon-by-Photon Analysis of Single-Molecule Fluorescence

On the single-molecule side, photon-by-photon single-molecule analysis is a practice developed by our group that, experimentally, records the arrival time of each detected photon and, theoretically, analyzes the photon time series using data-driven and model-free methods that are statistically robust. The change-point analysis is one such method developed by our lab. Further developments have extended this using Bayesian-inference based methods.


Super Line Localization

Much like super-resolution microscopy enables sub-diffraction limited resolution of single points, super-localization of line structures in noisy microscope images, which we have developed, enables the resolution of sub-diffraction limited line structures such as microtubules in cells.