journal article

Journal Article

Text

Neurocomputing. 2026, 670, p. 132563

Infinitely deep Bayesian neural network with signature transform

Elsevier B.V.

English

Text

2026-03-14

https://doi.org/10.1016/j.neucom.2025.132563

Yang X., Qiu P., Kawahara Y.. Infinitely deep Bayesian neural network with signature transform. Neurocomputing 670, 132563 (2026); https://doi.org/10.1016/j.neucom.2025.132563.

This work introduces a novel partially infinitely deep Bayesian neural network, particularly where the weights in each layer are governed by differential equations driven by the signature of Brownian motion. By leveraging the advantages of randomness compression for stochastic processes through signature transforms, our architecture is designed to overcome the limitations of current continuous deep Bayesian neural networks under stochasticity, particularly in terms of reliability and robustness. Additionally, we present a comprehensive mathematical framework that integrates the signature transform of stochasticity into the weight evolution of the Bayesian neural network. To approximate the true posterior, we adopt the approximate Bayesian computation method. Finally, empirical results across image classification tasks (CIFAR-10, CIFAR-10C) demonstrate that our model outperforms existing neural stochastic differential equation models in reliability and robustness while maintaining the memory-efficient training and accuracy of neural stochastic differential equations.

This article is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.

104018

Infinitely deep Bayesian neural network with signature transform

Elsevier B.V.

English

2026-03-14

Neurocomputing

670

132563

https://doi.org/10.1016/j.neucom.2025.132563

https://hdl.handle.net/11094/104018