Most modern learning theory starts with Bayes’ representation of knowledge from probability. In the 1970s, Andrei Kolmogorov put forth a different approach to statistics that coined Kolmogorov Complexity and how we solve the problem of data clustering.
Algorithmic statistics use the length in bytes of computer programs to measure how well that program (model, algorithm, etc.) fits your data. Rather than searching over a probability distribution, you search over a space of algorithms and settings. In this talk, we will explore the new cluster structure-function that tells us optimally how many clusters are in the dataset while it ensures that the differences within/between each cluster are as random and meaningful as possible.
We will discuss examples and results of such techniques: spatiotemporal human breast organoid patterning, proliferating cell applications, and plagiarism deterrence for beginning programmers.
Join on Zoom.