Introduction to Experimental Design

on YouTube ↗01h 04m

" summary

Covers adaptive experiment design under a fixed interventional Markov equivalence class, comparing worst-case, average-case and entropy-based greedy objectives and showing an explicit tree construction where information-greedy fails exponentially. When the essential graph is a tree, identifying the true DAG reduces to finding the root and a central-node bisection strategy recovers it in O(log K) single-node interventions and is within factor two of optimal. Also discusses targeted goals (shift interventions), batched experiments, and limits with latent confounding requiring multi-node interventions.

AI-written summary · about

" chapters (13)

1. 0:00Start
2. 0:48Greedy Strategies
3. 1:26Types of Greedy Strategies
4. 6:47The Information Greedy Strategy
5. 14:07Central Node Algorithm
6. 14:49Central Node
7. 24:20Targeted Experimental Design
8. 25:29Information Greedy Approach
9. 26:57Shift Intervention
10. 27:17Open Questions
11. 39:23Null Result
12. 46:01Lower Bounds
13. 46:36Single Node Interventions

" related talks

• 32mDesigning Adaptive Experiments For Policy Learning And InferenceMolly Offer-Westort · Learning from Interventions 2022
• 13mAlgorithmic Advances For The Design And Analysis Of Randomized Control TrialsChristopher Harshaw · Meet the Fellows Welcome Event
• 31mCost-effective data annotation with Bayesian experimental designNguyen Hong Quan · PyData Global 2024
• 40mApproximation Algorithms for Optimization under UncertaintyAnupam Gupta · Uncertainty in Computation 2016
• 39mModern Bayesian Experimental DesignDesi R. Ivavona · LAFI 2025
• 30mCounterfactual Inference For Sequential Experiment DesignRaaz Dwivedi · Learning from Interventions 2022
• 49mCausal Inference in Complex Systems: Network Interference, Strategic Agents, and BeyondPanos Toulis · Quantifying Uncertainty: Stochastic, Adversarial, and Beyond
• 29mSampling Algorithms for Combinatorial Pure ExplorationAnupam Gupta · Optimization, Statistics and Uncertainty