Extending Microservice Model Validity using Universal Differential Equations
Heimerson, Albin LU
and Ruuskanen, Johan
LU
(2023)
p.2401-2406
- Abstract
- When creating models of a system, there is always a tradeoff between the ease of modelling a part and the increased value it brings to the model. Learning a model using machine learning instead, we are able to capture all kinds of things we don't necessarily understand, but can need large amounts of data to learn even the things we find simple. Using universal differential equations we can combine the two, taking scientific models and embedding machine learning into them, with the goal of giving us the best of both worlds.
In this paper, we extend existing models with additional unmodelled dynamics using neural networks. We present results from a specific use-case involving a microservice fluid model, where the learned extension... (More) - When creating models of a system, there is always a tradeoff between the ease of modelling a part and the increased value it brings to the model. Learning a model using machine learning instead, we are able to capture all kinds of things we don't necessarily understand, but can need large amounts of data to learn even the things we find simple. Using universal differential equations we can combine the two, taking scientific models and embedding machine learning into them, with the goal of giving us the best of both worlds.
In this paper, we extend existing models with additional unmodelled dynamics using neural networks. We present results from a specific use-case involving a microservice fluid model, where the learned extension improves the range of parameters it can produce predictions over. This allows the optimization of the control parameters to be done to convergence in a single step, instead of collecting new data, in-between iterations, over multiple steps. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/d2715ce2-0507-48b7-9679-caa0a4934047
- author
- Heimerson, Albin
LU
and Ruuskanen, Johan
LU
- organization
- publishing date
- 2023
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- host publication
- IFAC-PapersOnLine
- pages
- 2401 - 2406
- external identifiers
-
- scopus:85183668294
- DOI
- 10.1016/j.ifacol.2023.10.1214
- project
- Event-Based Control of Stochastic Systems with Application to Server Systems
- language
- English
- LU publication?
- yes
- id
- d2715ce2-0507-48b7-9679-caa0a4934047
- date added to LUP
- 2023-03-06 14:31:37
- date last changed
- 2025-04-04 14:35:21
@inproceedings{d2715ce2-0507-48b7-9679-caa0a4934047,
abstract = {{When creating models of a system, there is always a tradeoff between the ease of modelling a part and the increased value it brings to the model. Learning a model using machine learning instead, we are able to capture all kinds of things we don't necessarily understand, but can need large amounts of data to learn even the things we find simple. Using universal differential equations we can combine the two, taking scientific models and embedding machine learning into them, with the goal of giving us the best of both worlds. <br/><br/>In this paper, we extend existing models with additional unmodelled dynamics using neural networks. We present results from a specific use-case involving a microservice fluid model, where the learned extension improves the range of parameters it can produce predictions over. This allows the optimization of the control parameters to be done to convergence in a single step, instead of collecting new data, in-between iterations, over multiple steps.}},
author = {{Heimerson, Albin and Ruuskanen, Johan}},
booktitle = {{IFAC-PapersOnLine}},
language = {{eng}},
pages = {{2401--2406}},
title = {{Extending Microservice Model Validity using Universal Differential Equations}},
url = {{https://lup.lub.lu.se/search/files/141037108/IFAC_2023_microservice_ude_No_template_.pdf}},
doi = {{10.1016/j.ifacol.2023.10.1214}},
year = {{2023}},
}