Generalized Exponential Concentration Inequality for Renyi Divergence Estimation

Authors: Shashank Singh, Barnabas Poczos

ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The main contribution of our work is to provide such a bound for an estimator of R enyi-α divergence for a smooth H older class of densities on the d-dimensional unit cube [0, 1]d. We also illustrate our theoretical results with a numerical experiment.
Researcher Affiliation Academia Shashank Singh SSS1@ANDREW.CMU.EDU Carnegie Mellon University, 5000 Forbes Ave., Pittsburgh, PA 15213 USA Barnab as P oczos BAPOCZOS@CS.CMU.EDU Carnegie Mellon University, 5000 Forbes Ave., Pittsburgh, PA 15213 USA
Pseudocode No The paper describes the estimation method mathematically but does not include any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any information or links regarding the availability of source code for the described methodology.
Open Datasets No We used our estimator to estimate the R enyi α-divergence between two normal distributions in R3 restricted to the unit cube. In particular, for p = N( µ1, Σ), q = N( µ2, Σ). For each n ∈ {1, 2, 5, 10, 50, 100, 500, 1000, 2000, 5000}, n data points were sampled according to each distribution and constrained (via rejection sampling) to lie within [0, 1]3. The paper generates synthetic data and does not use or provide access to a pre-existing public dataset.
Dataset Splits No For each n ∈ {1, 2, 5, 10, 50, 100, 500, 1000, 2000, 5000}, n data points were sampled according to each distribution and constrained (via rejection sampling) to lie within [0, 1]3. Our estimator was computed from these samples, for α = 0.8, using the Epanechnikov Kernel K(u) = 3 4(1 u2) on [ 1, 1], with constant bandwidth h = 0.25. The true α-divergence was computed directly according to its definition on the (renormalized) distributions on [0, 1]3. The bias and variance of our estimator were then computed in the usual manner based on 100 trials. The paper describes a simulation setup and data generation, but not explicit train/validation/test splits from a dataset.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running its experiments.
Software Dependencies No The paper does not provide specific software dependencies or versions (e.g., libraries with version numbers) used for the experiments.
Experiment Setup Yes Our estimator was computed from these samples, for α = 0.8, using the Epanechnikov Kernel K(u) = 3 4(1 u2) on [ 1, 1], with constant bandwidth h = 0.25.