Causal Inference For Socio-Economic And Engineering Systems

on YouTube ↗30m

" summary

Phase‑three trial data on 1,162 patients with ADAS‑Cog scores over two years and four arms (three therapies plus placebo) is framed as a missing‑entry tensor and imputed with tensor completion. A method called synthetic interventions fits PCA‑regularized linear models on donors’ pre‑intervention trajectories and applies those weights to donors’ post data; estimation uses principal component regression and relies on a low‑rank tensor, span and subspace inclusion, and selection‑on‑latent‑factors. Validation on 976 held‑out patients recreates post outcomes well and flags ~20% as potentially responsive; a patent is filed.

AI-written summary · about

" chapters (26)

1. 0:00Intro
2. 0:31Research agenda: Bridge causal inference/econometrics & machine learning
3. 1:15Bridging causal inference/econometrics & machine learning
4. 1:59Motivating application: Clinical trial study w. Alzheimer's Therapeutics company
5. 3:20Potential outcomes: a tensor viewpoint Potential outcomes
6. 4:08What data we had from
7. 4:45RCTS - randomization but no personalization
8. 5:29ADAS-COG score for each patient under each therapy? = Tensor completion
9. 6:52Bob and Alice: counterfactuals of interest
10. 7:22A partial answer: synthetic controls (SC) Abadie et al 03, '10
11. 9:01Synthetic controls - widely used but only a partial answer
12. 10:33A full answer: synthetic interventions Agarwal, Shah, Shen 21
13. 13:26Validation of Sl with Taux data: 976 held-out patients
14. 16:59What we found with SI
15. 18:22Additional case studies using synthetic interventions UBER Technologies
16. 18:49Essential questions
17. 19:00tensor factor model
18. 20:04linear span inclusion
19. 20:43What type of confounding is allowed?
20. 21:30selection on latent factors
21. 22:37subspace inclusion (i.e, when can you generalize?)
22. 23:28Putting it all together
23. 24:22Theoretical results
24. 26:23Analyzing the sample complexity of SI
25. 27:12Why PCR?
26. 29:06Summary

" related talks

• 11mCausal Inference For Socio-Economic & Engineering SystemsAnish Agarwal · Meet the Fellows Welcome Event
• 24mA Friendly Introduction To CausalityAlex Dimakis · MLconf SF 2016
• 31mA Gentle Introduction To Causal InferenceEyal Kazin · PyData Global 2022
• 30mKernel Methods For Causal InferenceRahul Singh · Algorithmic Aspects of Causal Inference 2022
• 01h 02mUnmeasured Confounding and More Recent Developments/Challenges in Causal DiscoveryDaniel Malinsky · Causality Boot Camp
• 31mPython and the Holy Grail of Causal InferenceDennis Ramondt, Huib Keemink · PyData Amsterdam 2018
• 40mActive Invariant Causal Prediction: Experiment Selection Through StabilityChristina Heinze-Deml · Algorithmic Aspects of Causal Inference 2022
• 20mInnovating Clinical Research with AI and Real World DataBenjamin Gmeiner · AI & Data Summit and Quantum Summit 2025