#### R250 Advanced Topics in ML and NLP # 6. Causal Inference <!-- Put the link to this slide here so people can follow --> Ferenc Huszár email: fh277@cam.ac.uk --- ## Introduction/Background 1. importance sampling 2. interventions and do-calculus 3. counterfactuals --- ## Paper discussions 1. importance sampling * [(Bottou et al, 2013)](https://www.microsoft.com/en-us/research/wp-content/uploads/2013/11/bottou13a.pdf) Counterfactual Reasoning in Learning Systems: The Example of Computational Advertising 1. interventions and do-calculus * [(Rezende et al, 2020)](https://arxiv.org/abs/2002.02836v1) Causally Correct Partial Models for Reinforcement Learning * additional [notes](https://www.inference.vc/notes-on-causally-correct-partial-models-2/) to help reading 1. counterfactuals * [(Kusner et al, 2017)](https://papers.nips.cc/paper/2017/file/a486cd07e4ac3d270571622f4f316ec5-Paper.pdf) Counterfactual Fairness --- ## Prerequisites * *assumed:* probabilistic modeling * conditional, joint, marginal distribution * expectations, conditional expectations * statistical independence * chain rule (probability), Bayes' rule * *helps if know a bit about*: * directed graphical models * importance sampling * basics of reinforcement learning --- ## Have questions? email: fh277@cam.ac.uk