Resource-Aware Time Series Imaging Classification for Wireless Link Layer Anomalies

Bertalanič, Blaž; Meža, Marko; Fortuna, Carolina

doi:10.1109/tnnls.2022.3149091

Cited by 18 publications

(8 citation statements)

References 36 publications

(42 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recently, the RP has been widely used in deep learning to transform a univariate time-series data into a 2-D image for CNN. It has been shown that the RP brings performance improvement compared to using the raw time series data in several applications [39], [40].…”

Section: Recurrence Plotmentioning

confidence: 99%

Multitask Deep Learning for Human Activity, Speed, and Body Weight Estimation Using Commercial Smart Insoles

Kim

Kang

Yang³

et al. 2023

IEEE Internet Things J.

View full text Add to dashboard Cite

Healthcare professionals and individual users use wearable devices equipped with various sensors for healthcare management. Recently, the joint usage of artificial intelligence and these wearable sensors has played an essential role in healthcare management by providing a wide range of applications such as fitness tracking, gym activity monitoring, patient rehabilitation monitoring, and disease detection. These tasks eventually aim to enhance personal well-being and better manage the user's physical health by monitoring different activity types and body weight changes. Here, we present an efficient multi-task learning framework based on commercial smart insoles that can solve three tasks related to physical health management: activity classification, speed estimation, and body weight estimation. Our multi-task framework converts the sensor data from the smart insole to a recurrence plot, which shows significant performance improvement compared to processing the raw time series data. In addition, we utilized a modified MobileNetV2 as our backbone network, which has a total parameter of less than 100K and a computational budget of 0.34G of multiply-accumulate operations. Furthermore, we collected a vast dataset from 72 users carrying out 16 experiments, which contains the largest number of people for multi-task learning purposes using smart insoles. Extensive experiments show that the proposed multi-task learning framework is extremely efficient while outperforming or leading to comparable performance against single-task models.

show abstract

Section: Recurrence Plotmentioning

confidence: 99%

Multitask Deep Learning for Human Activity, Speed, and Body Weight Estimation Using Commercial Smart Insoles

Kim

Kang

Yang³

et al. 2023

IEEE Internet Things J.

View full text Add to dashboard Cite

show abstract

“…The proposed classification model, that will serve as the backbone as pe Figure 1 for our TL model, is presented in Figure 3. Considering potential resource restrictions of BEMS the backbone model was developed according to the design consideration guidelines presented in [42], making the model resource-aware. During our iteration process we considered:…”

Section: A Proposed Modelmentioning

confidence: 99%

Dimensionality Expansion of Load Monitoring Time Series and Transfer Learning for EMS

Bertalanič¹,

Jenko²,

Fortuna³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

Electrical management systems (EMS) are playing a central role in enabling energy savings. They can be deployed within an everyday household where they monitor and manage appliances and help residents be more energy efficient and subsequently also more economical. One of they key functionalities of EMS is to automatically detect and identify appliances within a household through the process of load monitoring. In this paper, we propose a new transfer learning approach for building EMS (BEMS) and study the trade-offs in terms of numbers of samples and target classes in adapting a backbone model during the transfer process. We also perform a first time analysis of feature expansion through video-like transformation of time series data for device classification in non intrusive load monitoring (NILM) and propose a deep learning architecture enabling accurate appliance identification. We examine the relative performance of our method on 5 different representative low-frequency datasets and show that our method performs with an average F1 score of 0.88 on these datasets.

show abstract

“…Link-related anomalies caused by faulty hardware degradation or software related imperfections were observed by [10]. To automatically detect such faults, they proposed shallow ML-based techniques that were later proven to perform worse compared to DL-based methods [11]. However, these DL-based methods exhibited high computational complexity making them less suitable to run on constrained devices typical for DITEN setups.…”

Section: Introductionmentioning

confidence: 99%

Visibility Graph-Based Wireless Anomaly Detection for Digital Twin Edge Networks

Bertalanič,

Hribar,

Fortuna

2024

IEEE Open J. Commun. Soc.

Self Cite

View full text Add to dashboard Cite

Network softwarization, which shifts hardware-centric functions to software implementations, is essential for enhancing the agility of cellular and non-cellular wireless networks. This change, while raising reliability concerns, also improves system monitoring through digital twins. One example is the Digital Twin Edge Networks (DITEN), which enhances real-time analysis and quick anomaly response in the unpredictable last-mile wireless edge network environment. Positioned close to the physical infrastructure, DITEN is effective in rapidly detecting and addressing network irregularities. This study presents an advanced anomaly detection method for DITEN, employing time-series data conversion to Visibility Graph (VG) and utilising Graph Neural Network (GNN), with a focus on addressing disruptions at the network's physical layer. Our proposed method outperforms the State-of-the-art (SOTA) time series Deep Learning (DL) classification architecture by 13 percentage points and achieves ≈110 times higher computational efficiency. Furthermore, our method surpasses the classical Machine Learning (ML) model Hive-Cote2 by 2.2 percentage points while maintaining ≈5.9 times better computational efficiency. The model also outperforms the current best SOTA imaging model by up to 6 percentage points and the leading graph-based method by up to 10 percentage points, both with significantly lower Computational Complexity (CC) of ≈210-times and ≈4-times, respectively. Additionally, we show that when 1000 concurrent requests arrive, the proposed method achieves a mean response latency of less than or equal to 60 seconds across three setups. Finally, we demonstrate that the combination of Natural Visibility Graph (NVG) and the proposed GNN model provides interpretable insights by observing gradient changes.

show abstract

Resource-Aware Time Series Imaging Classification for Wireless Link Layer Anomalies

Cited by 18 publications

References 36 publications

Multitask Deep Learning for Human Activity, Speed, and Body Weight Estimation Using Commercial Smart Insoles

Multitask Deep Learning for Human Activity, Speed, and Body Weight Estimation Using Commercial Smart Insoles

Dimensionality Expansion of Load Monitoring Time Series and Transfer Learning for EMS

Visibility Graph-Based Wireless Anomaly Detection for Digital Twin Edge Networks

Contact Info

Product

Resources

About