Brief

Researchers have developed a new, mathematically sound, and computationally efficient method for measuring model complexity. This approach, based on analyzing similarities in model gradients across different inputs, is applicable to a wide range of models, including parametric, non-parametric, and kernel-based types. The proposed measure unifies and generalizes existing complexity metrics for various models like decision trees and neural networks, offering new insights into phenomena such as double descent. AI

Researchers have developed a new algorithm for learning mixture models that can handle heavy-tailed distributions, a significant improvement over previous methods that relied on low-degree moments. This novel approach utilizes efficient high-dimensional sparse Fourier transforms and does not require a minimum separation between cluster means, unlike algorithms for Gaussian mixtures. Additionally, a separate study introduces a regression framework that combines spectral representation learning with localized additive modeling to create interpretable models for heterogeneous data. AI