On the Suprema of Bernoulli Processes
summary
Proves the Bernoulli Conjecture: the expected supremum of a Bernoulli process is equivalent (up to universal constants) to the infimum over decompositions of T into two sets so that one part has bounded L1 norm and the other part is controlled by the Gaussian gamma2 functional; only those two mechanisms explain boundedness. Proof uses concentration, a Sudakov-type minoration for signs, and the new use of chopping maps and erasing variables. Applications include random Fourier series bounds and VC-class maximal inequalities.
AI-written summary · about
chapters (12)
- 0:00Introduction to suprema
- 3:33Gaussian processes
- 6:44Majorizing measures
- 11:59Bernoulli processes
- 19:06The Bernoulli conjecture
- 22:10Reverse bounds and tools
- 25:48Proof techniques
- 30:59Admissible partitions
- 34:55Chopping and erasing
- 38:04Practical applications
- 44:17VC classes and limits
- 47:20Empirical processes
