Potentials of Gated Higher Order Neural Units for Signal Decomposition and Process Monitoring

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F63839172%3A_____%2F25%3A10133656" target="_blank" >RIV/63839172:_____/25:10133656 - isvavai.cz</a>

Nalezeny alternativní kódy

RIV/68407700:21220/25:00383706 RIV/60076658:12310/25:43911223

Výsledek na webu

<a href="https://doi.org/10.1016/j.procs.2025.01.288" target="_blank" >https://doi.org/10.1016/j.procs.2025.01.288</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1016/j.procs.2025.01.288" target="_blank" >10.1016/j.procs.2025.01.288</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Potentials of Gated Higher Order Neural Units for Signal Decomposition and Process Monitoring

Popis výsledku v původním jazyce

In this paper, we propose an innovative shallow network architecture with gated higher-order neural units (gHONUs) for machine learning-based signal decomposition, which has the potential to improve anomaly detection and explainability while monitoring industrial processes as a real-time learning predictor. Our study shows that native HONU networks have limited decomposition capabilities due to the lack of a mechanism that forces the optimizer to enforce differential learning behavior among neurons. To address this limitation, we introduce pairs of neurons of different orders, connected by extended activation functions, as individual decomposition blocks. We present a first analysis of the behavior of these activation functions under HONU augmentations and show that the lowest linear correlation between paired neurons is achieved by a Tanh connection, where the neuron with an identity function output has a higher learning rate than the tanh signal constraining neuron. The possibilities of such a neural architecture are in applications of predictive and prescriptive maintenance and in analysis of time series resulting from complex dynamical behavior with incomplete observations and otherwise limited training, because the in-parameter linearity of nonlinear HONUs provides us with a mathematical-physical insight into the neural architecture.

Název v anglickém jazyce

Potentials of Gated Higher Order Neural Units for Signal Decomposition and Process Monitoring

Popis výsledku anglicky

In this paper, we propose an innovative shallow network architecture with gated higher-order neural units (gHONUs) for machine learning-based signal decomposition, which has the potential to improve anomaly detection and explainability while monitoring industrial processes as a real-time learning predictor. Our study shows that native HONU networks have limited decomposition capabilities due to the lack of a mechanism that forces the optimizer to enforce differential learning behavior among neurons. To address this limitation, we introduce pairs of neurons of different orders, connected by extended activation functions, as individual decomposition blocks. We present a first analysis of the behavior of these activation functions under HONU augmentations and show that the lowest linear correlation between paired neurons is achieved by a Tanh connection, where the neuron with an identity function output has a higher learning rate than the tanh signal constraining neuron. The possibilities of such a neural architecture are in applications of predictive and prescriptive maintenance and in analysis of time series resulting from complex dynamical behavior with incomplete observations and otherwise limited training, because the in-parameter linearity of nonlinear HONUs provides us with a mathematical-physical insight into the neural architecture.

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Projekt

<a href="/cs/project/LM2023054" target="_blank" >LM2023054: e-Infrastruktura CZ</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Rok uplatnění

2025

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

Procedia Computer Science

ISBN

—

ISSN

1877-0509

e-ISSN

—

Počet stran výsledku

10

Strana od-do

2278-2287

Název nakladatele

Elsevier B.V.

Místo vydání

Praha

Místo konání akce

Praha

Datum konání akce

13. 11. 2024

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

—