Efficient Distance Approximation for Structured High-Dimensional Distributions via Learning

Authors: Arnab Bhattacharyya, Sutanu Gayen, Kuldeep S Meel, N. V. Vinodchandran

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical We design efficient distance approximation algorithms for several classes of wellstudied structured high-dimensional distributions. Specifically, we present algorithms for the following problems (where d TV is the total variation distance): [...] All our algorithms are based on a common framework. To approximate the distance between P D1 and Q D2, we first learn the model parameters for P D1 and Q D2 that are guaranteed to be close to P and Q respectively. It remains to compute d TV(P, Q). This is a computationally hard problem in general, but we use the fact that for D1, D2 of interest, we can efficiently approximate the mass functions for P and Q from their parameters. At this point, we invoke an estimator that approximates d TV(P, Q) using samples from P and the approximate mass functions for P and Q.
Researcher Affiliation Academia Arnab Bhattacharyya National University of Singapore arnabb@nus.edu.sg Sutanu Gayen National University of Singapore sutanugayen@gmail.com Kuldeep S. Meel National University of Singapore meel@comp.nus.edu.sg N. V. Vinodchandran University of Nebraska-Lincoln vinod@cse.unl.edu
Pseudocode Yes Algorithm 1: Distance approximation between P and Q
Open Source Code No The paper does not provide any links to source code or explicitly state that source code for their methodology is made available.
Open Datasets No The paper is theoretical, presenting algorithms and analyses for different distribution types. It does not describe experiments using specific datasets, thus no information on public dataset availability is provided.
Dataset Splits No The paper is theoretical and does not describe experiments with datasets, thus no information on training/validation/test splits is provided.
Hardware Specification No The paper is theoretical and does not describe experimental setups, thus no hardware specifications are mentioned.
Software Dependencies No The paper is theoretical and does not describe experimental implementations with specific software dependencies and versions.
Experiment Setup No The paper is theoretical and focuses on algorithm design and analysis, not on concrete experimental setups or hyperparameter details.