Isabelle Guyon, President, ChaLearn
summary
ChaLearn organizes causality challenges that convert network reconstruction into concrete tests: a 2013 cause-effect pairs challenge supplies 20% real and 80% simulated pairs and attracts 267 participants. A winner uses massive feature generation plus random forests to reach AUC > 0.8 versus prior ~0.6, producing an operational orienting coefficient. For networks, network deconvolution is applied to inferred adjacency matrices to remove spurious indirect edges, proving near-perfect for linear models and effective but imperfect for nonlinear cases.
AI-written summary · about
chapters (19)
- 0:00Intro
- 0:23What is Big Data
- 2:12Data is all interconnected
- 5:07Questions are pervasive
- 5:23The scientific method
- 6:18Correlation
- 7:18Nobel Prize
- 7:30Correct Explanation
- 8:47Prioritize Experiments
- 9:38Machine Learning Literature
- 11:24Dependencies
- 12:34Removing dependencies
- 13:10Causeeffect Path Challenge
- 14:08Prime Setting
- 14:41Classification Problem
- 19:33Giving Data
- 20:23Results
- 22:27Network Reconstruction
- 22:48Network Deconvolution
related talks
24mA Friendly Introduction To Causality
29m🧠🧹 Causality - Mental Hygiene for Data Science
56mWhat can be passively learned about causality?
43mA Practical Guide to Causality in Python (For The Perplexed)
01h 02mUnmeasured Confounding and More Recent Developments/Challenges in Causal Discovery
28mLearning Causal Representations From Unknown Interventions
11mCausal Inference For Socio-Economic & Engineering Systems
01h 06mThe New Science of Cause and Effect