Extremes, outliers, and GOATS: on life in a lognormal world

on YouTube ↗30m

" summary

Claim: lifetime achievement often follows a lognormal rather than a Gaussian, so the top performer can be far ahead (Wayne Gretzky ~2900 vs ~1900 points). Two generative mechanisms are proposed: proportional gain (multiplicative yearly changes, simulation starts from Gaussian birth weights and yields lognormal adult weight) and weakest link (minimum of several multiplicative factors). Methods include fitting with SciPy tools, empiricaldist CDFs, minimizing area between curves, and notebooks that reproduce simulations and model comparisons; models mostly succeed.

AI-written summary · about

" chapters (14)

1. 0:00Introduction
2. 1:09Fast runners and GOATs
3. 3:18Why outliers exist
4. 4:22Proportional gain and lognormal
5. 8:42Central limit theorem corollary
6. 10:30Simulating weight distribution
7. 12:01Running speed as lognormal
8. 14:34Chess ratings and mechanisms
9. 15:58Gaussian talent vs achievement
10. 17:47Model comparison methodology
11. 21:38Code and notebooks
12. 25:37Summary and implications
13. 26:44Further reading
14. 27:05Q&A

" related talks

• 53mDesigning Robust LearnersJacob Steinhardt · Emerging Challenges in Deep Learning (Simons Institute workshop, 2019)
• 32mExtreme Value AnalysisDimitry Venger · PyData Tel Aviv 2022
• 34mInformation/Computation Gaps in Heavy-Tailed StatisticsSam Hopkins · Computational Phase Transitions (Simons Institute, 2020)
• 01h 23mImplicit Regularization IINati Srebro · Deep Learning Boot Camp 2019
• 48mComputational-Statistical Tradeoffs in Robust High-Dimensional EstimationIlias Diakonikolas · Robust and High-Dimensional Statistics (Simons Institute workshop, 2018)
• 48mLearning by Local Entropy MaximizationRiccardo Zecchina · Random Instances and Phase Transitions (Simons Institute, 2016)
• 01h 30mMaking Gaussian Processes UsefulBill Engels, Chris Fonnesbeck · PyData Global 2024
• 39mAbove Average-Case Analysis?Ankur Moitra · Learning, Algorithm Design and Beyond Worst-Case Analysis