InformationTheory

InformationTheory.Shannon.DPI

source

Data processing inequality for mutual information #

The data processing inequality I(X; f(Y)) ≤ I(X; Y) for deterministic post-processing.

Main statements #

Implementation notes #

klDiv_map_le is built via conditional Jensen (ConvexOn.map_condExp_le) applied to klFun and the Radon-Nikodym derivative identity Measure.rnDeriv_map. The mutual-information bound follows by applying klDiv_map_le to Prod.map id f.

theorem

InformationTheory.Shannon.klDiv_map_le

source

General pushforward DPI: klDiv (μ.map f) (ν.map f) ≤ klDiv μ ν for measurable f.

Used by
    theorem

    InformationTheory.Shannon.mutualInfo_le_of_postprocess

    source
    {Ω : Type u_1} [MeasurableSpace Ω] {X : Type u_2} [MeasurableSpace X] {Y : Type u_3} [MeasurableSpace Y] {Z : Type u_4} [MeasurableSpace Z] (μ : MeasureTheory.Measure Ω) [MeasureTheory.IsFiniteMeasure μ] (Xs : ΩX) (Yo : ΩY) (hXs : Measurable Xs) (hYo : Measurable Yo) {f : YZ} (hf : Measurable f) :
    mutualInfo μ Xs (f Yo) mutualInfo μ Xs Yo

    Data processing inequality: I(X; f(Y)) ≤ I(X; Y) for measurable f.

    Used by