Noise Quantification and Control in Circuits via Strong Data-Processing Inequalities
By: Chenyang Sun
Potential Business Impact:
Makes computers work even with errors.
This essay explores strong data-processing inequalities (SPDI's) as they appear in the work of Evans and Schulman \cite{ES} and von Neumann \cite{vN} on computing with noisy circuits. We first develop the framework in \cite{ES}, which leads to lower bounds on depth and upper bounds on noise that permit reliable computation. We then introduce the $3$-majority gate, introduced by \cite{vN} for the purpose of controlling noise, and obtain an upper bound on noise necessary for its function. We end by generalizing von Neumann's analysis to majority gates of any order, proving an analogous noise threshold and giving a sufficient upper bound for order given a desired level of reliability. The presentation of material has been modified in a way deemed more natural by the author, occasionally leading to simplifications of existing proofs. Furthermore, many computations omitted from the original works have been worked out, and some new commentary added. The intended audience has a rudimentary understanding of information theory similar to that of the author.
Similar Papers
Fundamental limits of distributed covariance matrix estimation via a conditional strong data processing inequality
Machine Learning (Stat)
Lets computers share data safely for better guessing.
Non-Linear Strong Data-Processing for Quantum Hockey-Stick Divergences
Quantum Physics
Makes secret messages safer in quantum computers.
Does the Data Processing Inequality Reflect Practice? On the Utility of Low-Level Tasks
Machine Learning (CS)
Makes computers learn better with messy data.