All Posts
On Problem Landscape and Conformal Prediction vs. Bayes Dilemma
- 07 July 2023
There are a lot of ongoing debates between the Bayesian approach and doing conformal prediction. And I think this debate should not exist at all. Conformal prediction is a great approach, and it is more or less orthogonal to the Bayesian analysis. I’ll try to explain my vision about how conformal prediction relates to Bayesian analysis, what they share in common and why they are still very different.
Bayes in the Wild
- 04 June 2023
I talk about the Bayesian approach to wide range of problems. Show how it is related to traditional methods in ML and what tasks benefit from an alternative view.
Bayesian AB Testing
- 21 April 2023
I talk about how Bayesian AB testing can drive conclusions from data. There is always the whole pipeline of decision making process: panning, execution and delivery. Each of the stages benefits from domain knowledge about the experimental setting. In the talk I explain how this can be framed from a Bayesian perspective.
Bayesian AB Testing
- 17 December 2022
I talk about the Bayesian approach to AB testing. The approach consists of 3 steps: making a hypothesis about the experiment, understanding time and data constraints, and interpreting the results after collecting the data. The advantage is that the Bayesian AB(C) test does not require p-values, corrections or bootstrap procedures, is conservative (does not exaggerate the result on small data) and is easily interpreted for business.
Bayesian inference at scale: Running A/B tests with millions of observations
- 12 August 2022
Industry data scientists are increasingly making the shift over to using Bayesian methods. However, one often cited reason for avoiding this is because “Bayesian methods are slow.” There can be some truth to this, although this is often a non-issue unless working at serious scale.
