2024
Klus, Roman; Talvitie, Jukka; Torres-Sospedra, Joaquín; Quezada-Gaibor, Darwin; Casteleyn, Sven; Cabric, Danijela; Valkama, Mikko
C2R: A Novel Architecture for Boosting Indoor Positioning With Scarce Data Journal Article
In: IEEE Internet of Things Journal, vol. 11, iss. 20, pp. 32868-32882, 2024, ISSN: 2327-4662, (2024-11).
Abstract | Links | BibTeX | Tags: deep learning, Indoor localization, Wi-Fi fingerprint
@article{Klus2025a,
title = {C2R: A Novel Architecture for Boosting Indoor Positioning With Scarce Data},
author = {Roman Klus and Jukka Talvitie and Joaquín Torres-Sospedra and Darwin Quezada-Gaibor and Sven Casteleyn and Danijela Cabric and Mikko Valkama},
doi = {https://doi.org/10.1109/JIOT.2024.3420122},
issn = {2327-4662},
year = {2024},
date = {2024-09-01},
urldate = {2024-09-01},
journal = {IEEE Internet of Things Journal},
volume = {11},
issue = {20},
pages = {32868-32882},
publisher = {IEEE},
abstract = {Improving the performance of Artificial Neural Network (ANN) regression models on small or scarce datasets, such as wireless network positioning data, can be realized by simplifying the task. One such approach includes implementing the regression model as a classifier, followed by a probabilistic mapping algorithm that transforms class probabilities into the multi-dimensional regression output. In this work, we propose the so-called c2r, a novel ANN-based architecture that transforms the classification model into a robust regressor, while enabling end-to-end training. The proposed solution can remove the impact of less likely classes from the probabilistic mapping by implementing a novel, trainable differential thresholded Rectified Linear Unit layer. The proposed solution is introduced and evaluated in the indoor positioning application domain, using 23 real-world, openly available positioning datasets. The proposed C2R model is shown to achieve significant improvements over the numerous benchmark methods in terms of positioning accuracy. Specifically, when averaged across the 23 datasets, the proposed c2r improves the mean positioning error by 7.9% compared to weighted knn with k=3, from 5.43 m to 5.00 m, and by 15.4% compared to a dense neural network (DNN), from 5.91 m to 5.00 m, while adapting the learned threshold. Finally, the proposed method adds only a single training parameter to the ann, thus as shown through analytical and empirical means in the article, there is no significant increase in the computational complexity.},
note = {2024-11},
keywords = {deep learning, Indoor localization, Wi-Fi fingerprint},
pubstate = {published},
tppubtype = {article}
}
Improving the performance of Artificial Neural Network (ANN) regression models on small or scarce datasets, such as wireless network positioning data, can be realized by simplifying the task. One such approach includes implementing the regression model as a classifier, followed by a probabilistic mapping algorithm that transforms class probabilities into the multi-dimensional regression output. In this work, we propose the so-called c2r, a novel ANN-based architecture that transforms the classification model into a robust regressor, while enabling end-to-end training. The proposed solution can remove the impact of less likely classes from the probabilistic mapping by implementing a novel, trainable differential thresholded Rectified Linear Unit layer. The proposed solution is introduced and evaluated in the indoor positioning application domain, using 23 real-world, openly available positioning datasets. The proposed C2R model is shown to achieve significant improvements over the numerous benchmark methods in terms of positioning accuracy. Specifically, when averaged across the 23 datasets, the proposed c2r improves the mean positioning error by 7.9% compared to weighted knn with k=3, from 5.43 m to 5.00 m, and by 15.4% compared to a dense neural network (DNN), from 5.91 m to 5.00 m, while adapting the learned threshold. Finally, the proposed method adds only a single training parameter to the ann, thus as shown through analytical and empirical means in the article, there is no significant increase in the computational complexity.
2022
Quezada-Gaibor, Darwin; Torres-Sospedra, Joaquín; Nurmi, Jari; Koucheryavy, Yevgeni; Huerta-Guijarro, Joaquín
SURIMI: Supervised Radio Map Augmentation with Deep Learning and a Generative Adversarial Network for Fingerprint-based Indoor Positioning Proceedings Article
In: 2022 IEEE 12th International Conference on Indoor Positioning and Indoor Navigation (IPIN), IEEE, 2022, ISBN: 978-1-7281-6218-8, (2022-35).
Abstract | Links | BibTeX | Tags: deep learning, Indoor positioning, machine learning
@inproceedings{Quezada2022d,
title = {SURIMI: Supervised Radio Map Augmentation with Deep Learning and a Generative Adversarial Network for Fingerprint-based Indoor Positioning},
author = {Darwin Quezada-Gaibor and Joaquín Torres-Sospedra and Jari Nurmi and Yevgeni Koucheryavy and Joaquín Huerta-Guijarro},
doi = {https://doi.org/10.1109/IPIN54987.2022.9918146},
isbn = {978-1-7281-6218-8},
year = {2022},
date = {2022-09-06},
urldate = {2022-09-06},
booktitle = {2022 IEEE 12th International Conference on Indoor Positioning and Indoor Navigation (IPIN)},
number = {1-8},
publisher = {IEEE},
abstract = {Indoor Positioning based on Machine Learning has drawn increasing attention both in the academy and the industry as meaningful information from the reference data can be extracted. Many researchers are using supervised, semi-supervised, and unsupervised Machine Learning models to reduce the positioning error and offer reliable solutions to the end-users. In this article, we propose a new architecture by combining Convolutional Neural Network (CNN), Long short-term memory (LSTM) and Generative Adversarial Network (GAN) in order to increase the training data and thus improve the position accuracy. The proposed combination of supervised and unsupervised models was tested in 17 public datasets, providing an extensive analysis of its performance. As a result, the positioning error has been reduced in more than 70% of them.},
note = {2022-35},
keywords = {deep learning, Indoor positioning, machine learning},
pubstate = {published},
tppubtype = {inproceedings}
}
Indoor Positioning based on Machine Learning has drawn increasing attention both in the academy and the industry as meaningful information from the reference data can be extracted. Many researchers are using supervised, semi-supervised, and unsupervised Machine Learning models to reduce the positioning error and offer reliable solutions to the end-users. In this article, we propose a new architecture by combining Convolutional Neural Network (CNN), Long short-term memory (LSTM) and Generative Adversarial Network (GAN) in order to increase the training data and thus improve the position accuracy. The proposed combination of supervised and unsupervised models was tested in 17 public datasets, providing an extensive analysis of its performance. As a result, the positioning error has been reduced in more than 70% of them.
Iskandaryan, Ditsuhi; Ramos-Romero, Francisco; Trilles-Oliver, Sergio
Application of deep learning and machine learning in air quality modeling Book Chapter
In: Marques, Gonçalo; Ighalo, Joshua (Ed.): pp. 11-23, Elsevier, 2022, ISBN: 9780323855976, (2022-15).
Links | BibTeX | Tags: air quality prediction, deep learning, machine learning
@inbook{Iskandaryan2022a,
title = {Application of deep learning and machine learning in air quality modeling},
author = {Ditsuhi Iskandaryan and Francisco Ramos-Romero and Sergio Trilles-Oliver},
editor = {Gonçalo Marques and Joshua Ighalo },
doi = {https://doi.org/10.1016/B978-0-323-85597-6.00018-5},
isbn = {9780323855976},
year = {2022},
date = {2022-03-30},
urldate = {2022-03-30},
pages = {11-23},
publisher = {Elsevier},
note = {2022-15},
keywords = {air quality prediction, deep learning, machine learning},
pubstate = {published},
tppubtype = {inbook}
}
2021
Belmonte-Fernández, Óscar; Sansano-Sansano, Emilio; Trilles-Oliver, Sergio; Caballer-Miedes, Antonio
In: Artificial Intelligence, Machine Learning, and Optimization Tools for Smart Cities, vol. 186, pp. 155-175, Springer, Cham, 2021, ISBN: 978-3-030-84459-2, (2021-21).
Abstract | Links | BibTeX | Tags: deep learning, Internet of things
@inbook{Belmonte2021a,
title = {A Reactive Architectural Proposal for Fog/Edge Computing in the Internet of Things Paradigm with Application in Deep Learning},
author = {Óscar Belmonte-Fernández and Emilio Sansano-Sansano and Sergio Trilles-Oliver and Antonio Caballer-Miedes},
doi = {https://doi.org/10.1007/978-3-030-84459-2_9},
isbn = {978-3-030-84459-2},
year = {2021},
date = {2021-09-01},
urldate = {2021-09-01},
booktitle = {Artificial Intelligence, Machine Learning, and Optimization Tools for Smart Cities},
volume = {186},
pages = {155-175},
publisher = {Springer, Cham},
series = {Springer Optimization and Its Applications},
abstract = {The fog/edge computing paradigm has been proposed to tackle the challenges inherent to the Internet of Things realm. Timely response, bandwidth efficiency, context awareness, data privacy and safety, and mobility support are some of the requirements that are only partially covered by cloud computing. A collaboration of both paradigms when developing deep learning solutions for the Internet of Things can be seen as a win–win approach. Time-consuming and hardware demanding deep learning models are built in the cloud with data provided by the fog/edge, and then these models are returned to the fog/edge for use. This work proposes a new architecture, based on the principles of reactive systems, for building responsive, resilient and elastic systems, where all components interact with one another through asynchronous message passing. As a proof of concept, two particular applications of this architecture in the realms of e-health and precision agriculture are presented.},
note = {2021-21},
keywords = {deep learning, Internet of things},
pubstate = {published},
tppubtype = {inbook}
}
The fog/edge computing paradigm has been proposed to tackle the challenges inherent to the Internet of Things realm. Timely response, bandwidth efficiency, context awareness, data privacy and safety, and mobility support are some of the requirements that are only partially covered by cloud computing. A collaboration of both paradigms when developing deep learning solutions for the Internet of Things can be seen as a win–win approach. Time-consuming and hardware demanding deep learning models are built in the cloud with data provided by the fog/edge, and then these models are returned to the fog/edge for use. This work proposes a new architecture, based on the principles of reactive systems, for building responsive, resilient and elastic systems, where all components interact with one another through asynchronous message passing. As a proof of concept, two particular applications of this architecture in the realms of e-health and precision agriculture are presented.