Bayesian Asymototics: From Universal Source Coding to General Large-Scale MIMO Capacity
Table of Contents
1. universal source coding
lossless source coding: known param => has optimal encoding; shannon limit
in practice: unknown param of distribution => encoder has a Coding distribution Q. => expected length = cross entropy
redundancy = different of Q and ideal Ptheta = KL div
Bayesian redundancy
minimax redundancy
jeffreys prior, jeffreys mixture, kt distribution, double mixture (context tree weight algoru=ithm)
1.1. connection to learning
good compressor = good predictor
2. Large-Scale MIMO Communication
non-linearity at receiver. general MIMO p(yreceiver|xtransmitter). focus nr->infinity
memoryless channel