Learning And Testing Causal Models: A Property Testing Viewpoint
summary
Frames causal testing as distribution property testing and gives tight sample trade-offs: trade-off between observational and interventional samples is explicit and information-theoretically optimal, with two-variable results requiring roughly K^C observational and min(√K, K^{(1-C)/2}) interventional samples and using unequal-sample closeness testers plus Levin’s work‑investment trick. For many variables with a known graph and bounded indegree/C‑component size, algorithms use O(log n) random interventions and polynomial samples per intervention via localization and approximate subadditivity.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 2:40Distribution property testing
- 4:10Property testing vs hypothesis testing
- 5:05Cancer and smoking
- 6:30interventional experiments
- 7:05how to realize this
- 15:05efficiency
- 15:40sample complexity
- 16:55setting
- 17:30causal causal networks
- 24:20localization
related talks
24mA Friendly Introduction To Causality
01h 01mThe Mathematics of Causal Inference, with Reflections on Machine Learning and the Logic of Science
39mCausal Effect Estimation in Practice: Lessons Learned from E-commerce & Banking
31mPython and the Holy Grail of Causal Inference
01h 11mStatistical Limits of Causal Inference
41mOn the Path to Causal Inference
49mCausal Inference in Complex Systems: Network Interference, Strategic Agents, and Beyond
31mA Gentle Introduction To Causal Inference