UAlberta Math Bio Seminar: Jay Newby

The data association problem is the difficult part of the multi-object tracking problem. The problem is to:

(i) classify measurements (e.g., position, shape, etc) as true positives or false positives,

(ii) classify tracked objects as observed (i.e., measured) or unobserved at each time, and

(iii) constructing a map (for each time step) from observed objects to true positive measurements.

I will share a recent breakthrough in applying MCMC methodology for approximating the distribution of tracking "hypotheses": the distribution over all possible tracks from a given dataset. I will illustrate the result with examples from tracking objects in microscopy videos, including salmonella, nuclei in fungal syncytia, and diatoms.