Beyond Normal: On the Evaluation of Mutual Information Estimators
This work addresses the need for better evaluation methods for mutual information estimators, which is crucial for practitioners in fields like representation learning and causality, though it is incremental in improving benchmarking practices.
The paper tackled the problem of evaluating mutual information estimators by constructing a diverse family of distributions with known ground-truth mutual information and proposing a benchmarking platform, revealing limitations of classical and neural estimators in high-dimensional and complex settings.
Mutual information is a general statistical dependency measure which has found applications in representation learning, causality, domain generalization and computational biology. However, mutual information estimators are typically evaluated on simple families of probability distributions, namely multivariate normal distribution and selected distributions with one-dimensional random variables. In this paper, we show how to construct a diverse family of distributions with known ground-truth mutual information and propose a language-independent benchmarking platform for mutual information estimators. We discuss the general applicability and limitations of classical and neural estimators in settings involving high dimensions, sparse interactions, long-tailed distributions, and high mutual information. Finally, we provide guidelines for practitioners on how to select appropriate estimator adapted to the difficulty of problem considered and issues one needs to consider when applying an estimator to a new data set.