An Information-Geometric Characterization of Chernoff Information
read more
Citations
An elementary introduction to information geometry
On a generalization of the Jensen-Shannon divergence and the JS-symmetrization of distances relying on abstract means.
Generalized Bhattacharyya and Chernoff upper bounds on Bayes error using quasi-arithmetic means ☆
An elementary introduction to information geometry
References
Elements of information theory
A Measure of Asymptotic Efficiency for Tests of a Hypothesis Based on the sum of Observations
Related Papers (5)
Frequently Asked Questions (5)
Q2. What is the function of the exponential family?
For distributions P1 and P2 of the same exponential family FF , indexed with respective naturalparameter θ1 and θ2, the α-Chernoff coefficient can be expressed analytically [12] as:cα(P1 : P2) = ∫ pα1 (x)p 1−α 2 (x)dν(x) = exp(−J (α) F (θ1 : θ2)), (5)where J (α)F (θ1 : θ2) is a skew Jensen divergence defined for F on the natural parameter space as:J (α) F (θ1 : θ2) = αF (θ1) + (1− α)F (θ2)− F (θ (α) 12 ), (6)where θ(α)12 = αθ1 + (1− α)θ2 = θ2 − α∆θ, with ∆θ = θ2 − θ1.
Q3. What is the definition of the expectation parameter?
4In convex analysis [13], each strictly convex and differentiable function F is associated with a dual convex conjugate F ∗ by the Legendre-Fenchel transformation: F ∗(η) = maxθ∈Θ〈η, θ〉−F (θ).
Q4. What is the function of the family FF?
Since log ∫ x∈X pθ(x)dν(x) = log 1 = 0, it follows that:F (θ) = − log ∫ exp(〈t(x), θ〉+ k(x))dν(x). (4)For full regular families [1], it can be proved that function F is strictly convex anddifferentiable over the open convex set Θ. Function F characterizes the family, and bears differentnames in the literature (partition function, log-normalizer or cumulant function) and parameterθ (natural parameter) defines the member Pθ of the family FF .
Q5. What is the function L defined over the Borel -algebra?
In practice, the authors often consider the Lebesguemeasure νL defined over the Borel σ-algebra E = B(Rd) of Rd for continuous distributionsFebruary 19, 2013 DRAFT4 (e.g., Gaussian), or the counting measure νC defined on the power set σ-algebra E = 2X for discrete distributions (e.g., Poisson or multinomial families).