Orthogonal Machine Learning: Power and Limitations

on YouTube ↗44m

" summary

Generalizes Neyman orthogonality to K-orthogonality and shows that if moment functions kill derivatives up to order K then nuisance estimators at rate N^(-1/(2K+2)) suffice for root-N inference; implements with sample-splitting and cross-fitting, estimates nuisances by LASSO in a partially linear regression, requires non-Gaussian treatment residuals (discrete discounts in pricing) for second‑order gains, and simulations (N≈5k, P≈1k) show much lower bias but slightly higher variance.

AI-written summary · about

" chapters (10)

1. 0:00Introduction
2. 3:59Classical Option
3. 10:59General Problem
4. 18:00Goal
5. 18:35First Order of Thought
6. 21:08Orthogonality
7. 21:44Results
8. 23:43In practice
9. 26:50Exploit nongaussian moments
10. 37:54Future Directions

" related talks

• 45mOrthogonal Statistical LearningVasilis Syrgkanis · Algorithmic Aspects of Causal Inference 2022
• 01h 22mGeneralization IVPeter Bartlett, Sasha Rakhlin · Deep Learning Boot Camp 2019
• 48mBenign Overfitting in Linear PredictionPeter Bartlett · Frontiers of Deep Learning 2019
• 33mThe Power and Limitations of Kernel LearningMisha Belkin · Optimization, Statistics and Uncertainty
• 59mTesting Assumptions of Learning AlgorithmsArsen Vasilyan · Sublinear Algorithms Boot Camp 2024
• 29mComputational Barriers to Estimation from Low-Degree PolynomialsTselil Schramm · Rigorous Evidence for Information-Computation Trade-offs
• 53mAdversarial Machine Learning and Instrumental Variables for Flexible Causal ModelingVasilis Syrgkanis · Adversarial Approaches in Machine Learning 2022
• 01h 02mPrediction, Generalization, Complexity: Revisiting the Classical View from Statistics Part 2Ryan Tibshirani · Modern Paradigms in Generalization Boot Camp 2024