High Dimensional Calibration for Rational Decision Making

on YouTube ↗29m

" summary

Provides an algorithm that produces d‑dimensional forecasts whose worst‑case bias grows like root t (up to a log factor in the number of events) and runs in time polynomial in the number of conditioning events so long as that number is polynomial in the problem size The method frames routing and online max‑weight matching as online combinatorial optimization and yields no‑regret guarantees on polynomially many subsequences, so published forecasts make best‑response behavior rational for many agents and apply to auctions, extensive‑form learning, and improved conformal prediction

AI-written summary · about

" chapters (10)

1. 0:00High dimensional calibration intro
2. 1:33Linear utility decision modeling
3. 4:45Calibration in binary/high dim
4. 7:34Calibration limitations
5. 11:15Online combinatorial framework
6. 16:20Subsequence regret analysis
7. 22:30Multi-agent low-bias predictions
8. 25:48Polynomial event prediction
9. 27:17Applications in mechanism design and machine learning
10. 29:07Summary

" related talks

• 01h 21mMulticalibration: Towards Fair Decision MakingMichael Kim · Data-Driven Decision Processes Boot Camp 2022
• 40mDecision-Making Under MiscalibrationGal Yona · Algorithmic Aspects of Causal Inference 2022
• 48mOnline Adversarial Multicalibration And (Multi)CalibeatingAaron Roth · Adversarial Approaches in Machine Learning 2022
• 33mOnline Multivalid Learning: Means, Moments, and Prediction IntervalsVarun Gupta, Aaron Roth, Mallesh M. Pai, Christopher Jung, Georgy Noarov · ITCS 2022
• 36mBeating the Curse of Dimensionality in High-Dimensional Optimal StoppingDavid Goldberg · Mathematics of Online Decision Making 2020
• 48mComputational-Statistical Tradeoffs in Robust High-Dimensional EstimationIlias Diakonikolas · Robust and High-Dimensional Statistics (Simons Institute workshop, 2018)
• 01h 08mData Driven Optimization Models and AlgorithmsYinyu Ye · Optimization, Statistics and Uncertainty
• 01h 07mRecent Progress in High-Dimensional LearningAnkur Moitra · Probability, Geometry, and Computation in High Dimensions Boot Camp (Simons Institute, Fall 2020)