Abstract: Imagine taking an aspirin for a pounding headache, only to have it cease right before you swallow the pill – was it the mere anticipation of relief or an arbitrary coincidence? Inferring causality is hard, and in this session we'll explore recent developments in the science of causal discovery, as well as motivate why this tool should be part of every data scientist's arsenal.
This is aimed at data scientists and industry practitioners with a working knowledge of basic statistics. We'll cover topics such as randomized control trials, conditional independence tests, causal discovery from observational data, and causal inference. We will show examples in code and provide real world examples of applications.
* Pearson to Pearl: A brief intro to causality and statistics
-- What are correlations,
-- Measuring Association vs measuring causal strength
-- Causal Graphs and structural causal equations
-- Structural causal equations
* Discovering causality
-- Conditional independence
-- The chicken and the egg problem
-- Basic causal discovery algorithms
-- Why should you care about causality?
-- Simpson's paradoxes everywhere
-- Asking counterfactual questions
* The future of AI
Basic knowledge of statistics, and a reasonable understanding of association measures
Bio: Andre joined causaLens from Goldman Sachs, where he was an executive director in the Model Risk Management group in Hong Kong and Frankfurt. Today he is working with industry leading, global organisations to apply cutting edge Causal AI research in production level solutions that empower individuals and teams to make better decisions. Andre received his PhD in theoretical physics from the University of Munich, where he studied the interplay between quantum mechanics and general relativity in black-holes.