Mateusz Smendowski
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Mateusz Smendowski has a total of 6 co-authors. Here are the top co-authors by number of shared publications:
- Piotr Nawrocki
Faculty of Computer Science, AGH University of Krakow, Krakow, Poland
8 shared publications - Nathalie Japkowicz
American University Washington DC USA
3 shared publications - Roberto Corizzo
Department of Computer Science, American University, Washington, USA
3 shared publications - Kamil Faber
AGH University of Krakow
2 shared publications - Leszek Rutkowski
Systems Research Institute of the Polish Academy of Sciences
1 shared publication - Mateusz Wojtulewicz
Faculty of Computer Science, AGH University of Krakow, Krakow, Poland
1 shared publication
Latest Publications
- Open Access
- 26 September 2026
- Published by Springer Nature in Machine Learning
- Vol. 115 (10) , 231
SPIRAL: A Novel Time Series to Image (TS2I) Transformation Method for Vision-Based Anomaly Detection
Time series anomaly detection (TSAD) underpins critical applications in manufacturing, healthcare, finance, and cloud computing. Recently, time series to image (TS2I) transformations have emerged as a promising approach that enables leveraging pretrained vision models for time series analysis. However, the majority of existing TS2I methods were designed for forecasting and classification, and suffer from parameter sensitivity, symmetric redundancy, and quadratic computational complexity. In this study, we introduce SPIRAL, a novel TS2I transformation designed for TSAD. SPIRAL maps time series windows onto a two-arm Archimedean spiral, producing asymmetric representations that preserve temporal locality while eliminating redundancy with linear time complexity with respect to window length. We additionally propose a standardized workflow with ACF-based window selection and point-wise scoring. Extensive evaluation (24,430 experiments) on 23 datasets from the TSB-AD benchmark, comparing SPIRAL against 9 TS2I transformations, 3 vision backbones, and 32 time-domain baselines across 9 metrics, shows that SPIRAL achieves the best average rank among 102 configurations in challenging metrics such as VUS-PR while exhibiting 40% lower training instability than its closest TS2I competitor. Our code is publicly available at: https://github.com/Smendowski/SPIRAL .
- 26 September 2026
- Published by Springer Nature in Machine Learning
- Vol. 115 (10) , 231
Open AccessSPIRAL: A Novel Time Series to Image (TS2I) Transformation Method for Vision-Based Anomaly Detection
Time series anomaly detection (TSAD) underpins critical applications in manufacturing, healthcare, finance, and cloud computing. Recently, time series to image (TS2I) transformations have emerged as a promising approach that enables leveraging pretrained vision models for time series analysis. However, the majority of existing TS2I methods were designed for forecasting and classification, and suffer from parameter sensitivity, symmetric redundancy, and quadratic computational complexity. In this study, we introduce SPIRAL, a novel TS2I transformation designed for TSAD. SPIRAL maps time series windows onto a two-arm Archimedean spiral, producing asymmetric representations that preserve temporal locality while eliminating redundancy with linear time complexity with respect to window length. We additionally propose a standardized workflow with ACF-based window selection and point-wise scoring. Extensive evaluation (24,430 experiments) on 23 datasets from the TSB-AD benchmark, comparing SPIRAL against 9 TS2I transformations, 3 vision backbones, and 32 time-domain baselines across 9 metrics, shows that SPIRAL achieves the best average rank among 102 configurations in challenging metrics such as VUS-PR while exhibiting 40% lower training instability than its closest TS2I competitor. Our code is publicly available at: https://github.com/Smendowski/SPIRAL .
- Open Access
- 1 September 2026
- Published by Springer Nature in Journal of Grid Computing
- Vol. 24 (3) , 25
Bridging Continual Learning and Green Cloud Computing: Foundations for Sustainable Time Series Anomaly Detection
While anomaly detection is essential for cloud computing and predictive maintenance, approaches that bridge continual learning with environmental sustainability remain largely unexplored. In production environments, evolving data distributions cause performance degradation of machine learning models, and naive adaptation may lead to forgetting of past patterns that are likely to reoccur. In this study, we formulate temporal concept-incremental anomaly detection for both univariate and multivariate time series. We introduce time-based and entity-based concept extraction strategies together with Wasserstein-distance diagnostics for inter- and intra-concept distribution drift, and define an evaluation protocol that measures detection quality, knowledge retention, and energy consumption. We further propose Selective Temporal Replay (STR), a replay strategy for autoencoder-based continual anomaly detection that combines diversity-based initialization, novelty-gated admission, retention-aware replacement, and age-weighted preservation of stored sequences. The empirical study comprises more than 5,400 experiments across four datasets, nine autoencoder architectures, and ten continual learning strategies. On Yahoo! A1, STR obtains the highest ROC-AUC in seven of the nine architectures and is the only strategy with positive backward transfer on all four datasets; under the smallest memory budgets, it obtains the highest ROC-AUC among replay methods on Yahoo! A1 and SMD. The sustainability analysis identifies when selective replay justifies its additional energy cost and when regularization-based continual learning strategies are preferable.
- 1 September 2026
- Published by Springer Nature in Journal of Grid Computing
- Vol. 24 (3) , 25
Open AccessBridging Continual Learning and Green Cloud Computing: Foundations for Sustainable Time Series Anomaly Detection
While anomaly detection is essential for cloud computing and predictive maintenance, approaches that bridge continual learning with environmental sustainability remain largely unexplored. In production environments, evolving data distributions cause performance degradation of machine learning models, and naive adaptation may lead to forgetting of past patterns that are likely to reoccur. In this study, we formulate temporal concept-incremental anomaly detection for both univariate and multivariate time series. We introduce time-based and entity-based concept extraction strategies together with Wasserstein-distance diagnostics for inter- and intra-concept distribution drift, and define an evaluation protocol that measures detection quality, knowledge retention, and energy consumption. We further propose Selective Temporal Replay (STR), a replay strategy for autoencoder-based continual anomaly detection that combines diversity-based initialization, novelty-gated admission, retention-aware replacement, and age-weighted preservation of stored sequences. The empirical study comprises more than 5,400 experiments across four datasets, nine autoencoder architectures, and ten continual learning strategies. On Yahoo! A1, STR obtains the highest ROC-AUC in seven of the nine architectures and is the only strategy with positive backward transfer on all four datasets; under the smallest memory budgets, it obtains the highest ROC-AUC among replay methods on Yahoo! A1 and SMD. The sustainability analysis identifies when selective replay justifies its additional energy cost and when regularization-based continual learning strategies are preferable.
- 25 August 2026
- Published by Springer Nature
- p. 39-55
PRISM: Powerful Time Series to Image (TS2I) Representations for Multivariate Anomaly Detection
- 25 August 2026
- Published by Springer Nature
- p. 39-55
PRISM: Powerful Time Series to Image (TS2I) Representations for Multivariate Anomaly Detection
- 9 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Cloud Computing
- Vol. 14 (2) , 731-743
A Novel Balanced Binary Whale Optimization Algorithm for Dynamic Feature Selection in Green Cloud Computing
This paper introduces a novel Balanced Binary Whale Optimization Algorithm (BB-WOA) designed specifically for dynamic feature selection in Green Cloud Computing (GCC). Traditional feature selection methods used in cloud resource forecasting often suffer from either suboptimal predictive performance or excessive computational complexity. To address this, we propose significant algorithmic enhancements over the standard Binary Whale Optimization Algorithm (B-WOA), including dynamic binary transition functions, progressive scaling, diversified population initialization via Sobol sequences, balanced exploration-exploitation strategies, and an activation-based recovery mechanism. Our comprehensive experimental evaluation using real-world cloud resource data demonstrates that BB-WOA outperforms existing methods. Specifically, BB-WOA reduces predictive error (RMSE) by up to 7.45% compared to B-WOA and 1.55% compared to Genetic Algorithm (GA). Moreover, BB-WOA achieves computational improvements, reducing execution time by approximately 38.30%, 64.13%, and 78.53% over B-WOA, Random Search (RS), and GA, respectively. Environmentally, the proposed method reduces energy consumption by 38.48% relative to B-WOA, 64.18% compared to RS, and 52.55% relative to GA, while simultaneously selecting significantly fewer features (a reduction of up to 70.77%). These results underscore the effectiveness and sustainability of BB-WOA, positioning it as a highly competitive and environmentally friendly solution.
- 9 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Cloud Computing
- Vol. 14 (2) , 731-743
A Novel Balanced Binary Whale Optimization Algorithm for Dynamic Feature Selection in Green Cloud Computing
This paper introduces a novel Balanced Binary Whale Optimization Algorithm (BB-WOA) designed specifically for dynamic feature selection in Green Cloud Computing (GCC). Traditional feature selection methods used in cloud resource forecasting often suffer from either suboptimal predictive performance or excessive computational complexity. To address this, we propose significant algorithmic enhancements over the standard Binary Whale Optimization Algorithm (B-WOA), including dynamic binary transition functions, progressive scaling, diversified population initialization via Sobol sequences, balanced exploration-exploitation strategies, and an activation-based recovery mechanism. Our comprehensive experimental evaluation using real-world cloud resource data demonstrates that BB-WOA outperforms existing methods. Specifically, BB-WOA reduces predictive error (RMSE) by up to 7.45% compared to B-WOA and 1.55% compared to Genetic Algorithm (GA). Moreover, BB-WOA achieves computational improvements, reducing execution time by approximately 38.30%, 64.13%, and 78.53% over B-WOA, Random Search (RS), and GA, respectively. Environmentally, the proposed method reduces energy consumption by 38.48% relative to B-WOA, 64.18% compared to RS, and 52.55% relative to GA, while simultaneously selecting significantly fewer features (a reduction of up to 70.77%). These results underscore the effectiveness and sustainability of BB-WOA, positioning it as a highly competitive and environmentally friendly solution.
- Top Cited Papers
- 8 January 2025
- Published by Springer Nature in Journal of Grid Computing
- Vol. 23 (1) , 1-35
A Survey of Cloud Resource Consumption Optimization Methods
- 8 January 2025
- Published by Springer Nature in Journal of Grid Computing
- Vol. 23 (1) , 1-35
Top Cited PapersA Survey of Cloud Resource Consumption Optimization Methods
- Open Access
- 7 September 2024
- Published by Elsevier in Knowledge-Based Systems
- Vol. 304
Optimizing multi-time series forecasting for enhanced cloud resource utilization based on machine learning
- 7 September 2024
- Published by Elsevier in Knowledge-Based Systems
- Vol. 304
Open AccessOptimizing multi-time series forecasting for enhanced cloud resource utilization based on machine learning
- Open Access
- 1 July 2024
- Published by Walter de Gruyter GmbH in Journal of Artificial Intelligence and Soft Computing Research
- Vol. 14 (4) , 287-308
Optimization of the Use of Cloud Computing Resources Using Exploratory Data Analysis and Machine Learning
Rapid growth in the popularity of cloud computing has been largely caused by increasing demand for scalable IT solutions, which could provide a cost-effective way to manage the software development process and meet business objectives. Optimization of cloud resource usage remains a key issue given its potential to significantly increase efficiency and flexibility, minimize costs, ensure security, and maintain high availability of services. This paper presents a novel concept of a Cloud Computing Resource Prediction and Optimization System, which is based on exploratory data analysis that acknowledges, among others, the information value of outliers and dynamic feature selection. The optimization of cloud resource usage relies on long-term forecasting, which is considered a dynamic and proactive optimization category. The analysis presented here focuses on the applicability of classical statistical models, XGBoost, neural networks and Transformer. Experimental results reveal that machine learning methods are highly effective in long-term forecasting. Particularly promising results – in the context of potential prediction-based dynamic resource reservations – have been yielded by prediction methods based on the BiGRU neural network and the Temporal Fusion Transformer.
- 1 July 2024
- Published by Walter de Gruyter GmbH in Journal of Artificial Intelligence and Soft Computing Research
- Vol. 14 (4) , 287-308
Open AccessOptimization of the Use of Cloud Computing Resources Using Exploratory Data Analysis and Machine Learning
Rapid growth in the popularity of cloud computing has been largely caused by increasing demand for scalable IT solutions, which could provide a cost-effective way to manage the software development process and meet business objectives. Optimization of cloud resource usage remains a key issue given its potential to significantly increase efficiency and flexibility, minimize costs, ensure security, and maintain high availability of services. This paper presents a novel concept of a Cloud Computing Resource Prediction and Optimization System, which is based on exploratory data analysis that acknowledges, among others, the information value of outliers and dynamic feature selection. The optimization of cloud resource usage relies on long-term forecasting, which is considered a dynamic and proactive optimization category. The analysis presented here focuses on the applicability of classical statistical models, XGBoost, neural networks and Transformer. Experimental results reveal that machine learning methods are highly effective in long-term forecasting. Particularly promising results – in the context of potential prediction-based dynamic resource reservations – have been yielded by prediction methods based on the BiGRU neural network and the Temporal Fusion Transformer.
- 27 April 2024
- Published by Elsevier in Journal of Computational Science
- Vol. 79
FinOps-driven optimization of cloud resource usage for high-performance computing using machine learning
- 27 April 2024
- Published by Elsevier in Journal of Computational Science
- Vol. 79
FinOps-driven optimization of cloud resource usage for high-performance computing using machine learning