Neural Mutual Information Estimation with Vector Copulas
This addresses the fundamental task of mutual information estimation in data science and machine learning, offering a more balanced approach for practitioners dealing with complex distributions.
The paper tackled the problem of estimating mutual information by proposing a method that interpolates between flexible neural models and simplified Gaussian copulas to achieve a better trade-off between complexity and capacity, demonstrating advantages on synthetic benchmarks and real-world data.
Estimating mutual information (MI) is a fundamental task in data science and machine learning. Existing estimators mainly rely on either highly flexible models (e.g., neural networks), which require large amounts of data, or overly simplified models (e.g., Gaussian copula), which fail to capture complex distributions. Drawing upon recent vector copula theory, we propose a principled interpolation between these two extremes to achieve a better trade-off between complexity and capacity. Experiments on state-of-the-art synthetic benchmarks and real-world data with diverse modalities demonstrate the advantages of the proposed estimator.