Defines the Gaussian entropy pair on the noise Wasserstein space: the penalty is beta times relative entropy with respect to the diagonal Gaussian reference measure, and the score is beta times the noise score field on measures of finite weighted Fisher information.
In the setting of A Diagonal Gaussian Reference Measure on the Noise Wasserstein Space, Rescaled Heads and Gaussian Tails: Standing Notation, so that the reference measure is , let be positive.
1. (Hypothesis) There is a positive with for every .
2. (The penalty domain) is the set of the that have finite relative entropy with respect to , so that is a real number for . By the hypothesis of clause 1, by The Entropy Domain of a Diagonal Gaussian Reference Measure Has the Noise Map Property §inclusion; and , the set of The Second Moment of a Borel Probability Measure on a Hilbert Space and the Probability Measures with Finite Second Moment §space, by Relative Entropy with Respect to a Diagonal Gaussian Measure on a Hilbert Space: the Moment Bound, the Cutoff Projections, and Bounded, Tight, Weakly Closed, Wasserstein-Closed and Weakly Sequentially Compact Sublevel Sets §moment.
3. (The score domain) is the set of the that have a relative score with respect to and finite Fisher information relative to with weights , these notions being applicable since by clause 2. For , is the noise score field of , which lies in the noise tangent space by The Noise Score Field of a Measure of Finite Weighted Fisher Information: Existence, Norm, Pairing with Noise Gradients, Head Approximation and Tangency §tangent; since is a linear subspace of by Linearity of the Noise Gradient, and the Noise Tangent Space is a Closed Linear Subspace §subspace, also .
4. (The Gaussian entropy pair) Under the hypothesis of clause 1, the Gaussian entropy pair with temperature is the quadruple consisting of the sets of clauses 2 and 3 and the functions
Loading…
No relations recorded yet.