Journal DOI 10.68956/journal.human-performance-futures

Creative Commons source reading · Prediction and forecasting

An Introduction to Stochastic Deep Learning

Faming Liang

Originally published in Wiley Interdiscip Rev Comput Stat · 2026-09-19

From the article

Selected source text. JournalOS removes markup and shortens excerpts where indicated. The full article provides the complete context.

Abstract

Deep neural networks (DNNs) have achieved remarkable success in prediction, but their deterministic formulation makes many statistical inference tasks difficult. StoNet, short for stochastic neural network, addresses this limitation by reformulating a DNN as a probabilistic latent-variable model, in which the outputs of selected hidden layers or units are treated as latent variables. The article describes the formulation of StoNet and its asymptotic equivalence to a conventional DNN, showing how StoNet can serve both as a deep learning model and as an analytical device for studying DNNs. It reviews training algorithms for StoNet and discusses its applications to statistical inference problems, including nonlinear sufficient dimension reduction, causal inference with observed and unobserved confounders, Granger-causality learning for…

Conclusion

This article provides a concise review of StoNet, covering its formulation, theoretical foundations, computational methods, and applications. StoNet and its variants—including Causal-StoNet, K-StoNet, sparse StoNet heads, and sublinearly structured DNNs—demonstrate how probabilistic modeling and deep learning can be integrated to address fundamental inferential challenges in modern data science. Table 1 summarizes the main applications and extensions discussed in this article, together with their computational methods, available software implementations, and references. The central message is that, when stochasticity is introduced through a coherent latent-variable structure, deep neural networks become more statistically analyzable: their internal structures are more interpretable, and their prediction uncertainty is more…

Editorial reading guide

  • What data created and tested the model, and was evaluation independent?
  • What is the cost of a false positive, false negative or out-of-distribution decision?