Bickel's Influence On My Career
summary
Three concrete claims and how they are proved: covariance estimation shows that for banding models a sample size of order log P suffices and spectral-norm rates come from blockwise analysis with optimal K choice; for sparse covariance the procedure is minimax when S < sqrt(n) optimality via a new lower-bound combining Le Cam and Fano ideas; network work gives method-of-moments consistency at 1/n and, for the realized community labels, a 1/n^2 rate; variational inference finds that mean-field variational Bayes works under strong SNR with coordinate-ascent/EM refinements and provable local-to-global guarantees.
AI-written summary · about