To overcome this challenge, diverging from conventional efforts of expanding the solver's search scope, we focus on enabling information to actively propagate to the solver through heat diffusion.

Artificial neural networks are among the most successful modern machine learning methods, in particular because their non-linear parametrization provides a flexible way to implement feature learning (see, e.g., Goodfellow et al., 2016, chapter 15).

Molecular Property Prediction (MPP) is a critical task in computational drug discovery, aimed at identifying molecules with desirable pharmacological and ADMET (absorption, distribution, metabolism, excretion, and toxicity) properties. Machine learning models have been widely used in this fast-growing field, with two types of models being commonly employed: traditional non-deep models and deep models.

However, only the smooth setting was considered. To close this gap, we introduce and analyze O-ZD, the first structured finite-difference algorithm for non-smooth black-box optimization.

Temporal sequence processing is fundamental in brain cognitive functions.

Finding tight linear bounds for activation functions in neural networks is an essential part of several state of the art neural network robustness certification tools.

Neural Information Processing Systems http://nips.cc/

The fundamental learning theory behind neural networks remains largely open. What classes of functions can neural networks actually learn?