Created:
Generative models learn to map noise distributions to data distributions. This is a paradigm shift from the single predictions to sampling from a learned distribution, and enables fun things to explore like solving inverse problems.
Created:
Many master students I know are applying for doctoral programs now, which makes me want to write down some of my thoughts. I won’t talk about commonly discussed topics. Instead, I want to talk about training in an industry-relevant field.
Created:
Computing moving statistics (e.g. mean and variance) is useful when dealing with very large datasets. Welford’s algorithm is one method for achieving this. This post derives how it works for single sample as well as batch updates.
Created:
The resources/documentations out there can be overwhelming for scientific/academic software development. The general concept of building, managing dependencies, and CI/CD might be gently introduced/reviewed in this lecture, but one could probably find resources more updated and specific to Python elsewhere.
Created:
This perspective describes existing algorithms for scaling and merging in crystallography and the recent introduction of variational inference (VI) in this field.