Pith. sign in

REVIEW 2 cited by

A Benchmark Suite for Evaluating Neural Mutual Information Estimators on Unstructured Datasets

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2410.10924 v1 pith:JPCXWUMD submitted 2024-10-14 stat.ML cs.AIcs.LG

A Benchmark Suite for Evaluating Neural Mutual Information Estimators on Unstructured Datasets

classification stat.ML cs.AIcs.LG
keywords datasetsestimatorsbenchmarkneuralsuiteunstructuredanalyticalevaluating
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Mutual Information (MI) is a fundamental metric for quantifying dependency between two random variables. When we can access only the samples, but not the underlying distribution functions, we can evaluate MI using sample-based estimators. Assessment of such MI estimators, however, has almost always relied on analytical datasets including Gaussian multivariates. Such datasets allow analytical calculations of the true MI values, but they are limited in that they do not reflect the complexities of real-world datasets. This study introduces a comprehensive benchmark suite for evaluating neural MI estimators on unstructured datasets, specifically focusing on images and texts. By leveraging same-class sampling for positive pairing and introducing a binary symmetric channel trick, we show that we can accurately manipulate true MI values of real-world datasets. Using the benchmark suite, we investigate seven challenging scenarios, shedding light on the reliability of neural MI estimators for unstructured datasets.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Towards Diverse and Comprehensive Benchmarks for Mutual Information Estimation

    cs.LG 2026-07 accept novelty 6.5

    A copula-theoretic benchmark suite reveals that non-parametric, discriminative and generative MI estimators each dominate only in specific regimes, with no universal winner.

  2. Estimating classical mutual information between quantum subsystems with neural networks

    quant-ph 2025-08 unverdicted novelty 5.0

    Neural networks reconstruct classical mutual information and specific entropy from limited projective measurements in the antiferromagnetic quantum Ising model, enabling reconstruction of the phase diagram even for de...