In this paper, we propose a novel lossless data compression framework, called Stochastic Pattern Discrete Flow (SPDFlow), designed for lightweight IoT applications. SPDFlow integrates normalizing flow models with range Asymmetric Numeral Systems (rANS) to efficiently compress high-frequency and stochastic time-series smart metering data. Building on advances in invertible flow-based distribution mapping, we design an online compression scheme optimized for electricity usage profiles sampled from tens of milliseconds to several seconds. The proposed linear and lightweight architecture enables real-time deployment on resource-constrained edge devices while maintaining robust modeling capability under unpredictable, event-driven peaks. A customized discretization strategy allows efficient encoding of wide-ranging measurement values. Comprehensive theoretical analysis and experiments across diverse datasets verify SPDFlow’s superior compression performance compared with both traditional and recent learning-based methods achieving up to 31% data volume reduction under short update intervals. Furthermore, real-world deployment on Raspberry Pi devices confirms its practical feasibility. To the best of our knowledge, this is the first work to explore lossless compression of time-series load profiles using normalizing flows.

ClinicalTrials.gov Identifier: NCT02579499.

In this paper, we propose a novel lossless data compression framework, called Stochastic Pattern Discrete Flow (SPDFlow), designed for lightweight IoT applications. SPDFlow integrates normalizing flow models with range Asymmetric Numeral Systems (rANS) to efficiently compress high-frequency and stochastic time-series smart metering data. Building on advances in invertible flow-based distribution mapping, we design an online compression scheme optimized for electricity usage profiles sampled from tens of milliseconds to several seconds. The proposed linear and lightweight architecture enables real-time deployment on resource-constrained edge devices while maintaining robust modeling capability under unpredictable, event-driven peaks. A customized discretization strategy allows efficient encoding of wide-ranging measurement values. Comprehensive theoretical analysis and experiments across diverse datasets verify SPDFlow’s superior compression performance compared with both traditional and recent learning-based methods achieving up to 31% data volume reduction under short update intervals. Furthermore, real-world deployment on Raspberry Pi devices confirms its practical feasibility. To the best of our knowledge, this is the first work to explore lossless compression of time-series load profiles using normalizing flows.

ClinicalTrials.gov Identifier: NCT02579499.

Sam Jin Pharmaceutical, Seoul, Korea and Chong Kun Dang Pharmaceutical, Seoul, Korea.

Accurate prediction of Remaining Useful Life (RUL) is critical for effective prognostics and health management (PHM) in industrial systems. However, existing data-driven methods often require large amounts of high-quality data and lack interpretability, limiting their applicability in real-world scenarios. In practice, widely used synthetic benchmark datasets such as CMAPSS and N-CMAPSS are themselves noisy due to sensor drift and mode switching, and limited in the number of available run-to-failure trajectories. In this study, we propose ST-KDFormer (Spatio-Temporal Koopman Dual-branch Transformer), a novel architecture that integrates Koopman operator theory with deep learning to address these challenges. ST-KDFormer employs separate Koopman Autoencoders (KAEs) to learn linearized latent representations of temporal and spatial system dynamics. These embeddings are jointly processed through a dual-branch Transformer encoder to capture contextual dependencies within and across temporal windows and sensor dimensions. To enable efficient end-to-end training, we introduce a unified loss function with adaptive weighting to balance embedding and RUL prediction objectives. We evaluate our approach on the CMAPSS and N-CMAPSS benchmark datasets through extensive experiments, including ablation studies and comparisons with state-of-the-art methods. The results demonstrate that ST-KDFormer achieves superior performance in terms of prediction accuracy and robustness across varying operational conditions, validating its effectiveness for practical RUL estimation tasks.

In this study, we propose a novel method for predicting the number of household members using smart meter electricity consumption data. Utilizing various feature selection techniques, including SHAP values, mutual information, feature importance, and a proposed method involving pairwise grouping and selection based on importance testing, we aimed to enhance the model's predictive accuracy. The proposed method involves selecting time points based on specific time ranges and overlapping important features from other techniques. The analysis revealed that time periods such as morning, evening, and before bedtime are most informative for estimating household member numbers, while periods with lower activity, like lunchtime to early evening, are less useful.

In this paper, we present a novel key-indexed channel phase permutation method designed to enhance the security of physical-layer (PHY) authentication in the presence of correlated sub-channels. In wireless communication, secret keys should be protected from eavesdroppers, as unauthorized access to these keys may compromise user security, expose sensitive information, and allow attackers to impersonate legitimate users, bypassing authentication mechanisms. To avoid the malicious interception of the pre-shared secret keys in wireless communication, the proposed scheme takes advantage of the inherent characteristics of reciprocal radio frequency channel phases and the key indices between authorized users. This ensures the completion of PHY authentication process at the PHY layer without divulging the actual secret key to potential attackers. Channel phase data are permuted based on binary key indices and concealed within phase information, thwarting unauthorized access. Through extensive testing using a Universal Software Radio Peripheral testbed, we validate the robustness of our scheme to various attacks, ensuring the confidentiality of secret keys and distinguishing legitimate users from attackers attempting illicit authentication. This method strengthens PHY layer security, offering practical applicability in wireless communication scenarios.

In this paper, we address the problem of non-robustness in empirical risk minimization-based learning schemes for indoor localization using radio frequency (RF) fingerprints. To overcome this challenge, we propose a novel solution that leverages the manifold structure of the channel state information (CSI) data, exploiting its geographical similarity. Our approach involves a simple yet effective data augmentation scheme named Geomixup, which utilizes geographic information to synthesize data for training. By incorporating geolocation-based data aug-mentation, we enable robust localization based on CSI, ensuring that the model's decisions are grounded in the highly correlated geolocation feature, rather than weakly correlated ones. To evaluate the performance of the proposed method, we utilized two performance measures, pseudo accuracy and distance error, on the data points that were not encountered during training. They consider the confidence assigned to each location prediction and apply a larger penalty in cases where the model exhibits high confidence in misplacement. Through extensive RF - based localization tests, we demonstrate that our proposed scheme outperforms other learning methods in terms of robustness, all while maintaining a negligible loss of accuracy. This significant improvement in robustness ensures the reliability and stability of our approach, making it highly suitable for real-world indoor localization applieations.

The rapid growth of the mHealth market has led to the development of several tools to evaluate user experience. However, there is a lack of universal tools specifically designed for this emerging technology. This study was conducted with the aim of developing and verifying a user experience evaluation scale for mHealth apps based on factors proposed in previous research. The initial draft of the tool was created following a comprehensive review of existing questionnaires related to mHealth app evaluation. The validity of this scale was then tested through exploratory and confirmatory factor analysis. The results of the factor analysis led to the derivation of 16 items, which were conceptually mapped to five factors: ease of use and satisfaction, information architecture, usefulness, ease of information, and aesthetics. A case study was also conducted to improve mHealth apps concerning personal health records using this scale. In conclusion, the developed user experience evaluation scale for mHealth apps can provide comprehensive user feedback and contribute to the improvement of these apps.

With global EV sales projected to reach 3.5 million units by 2023 and public charging stations increasing by 40%, effective management and optimization of charging infrastructure have become critical. While current research primarily focuses on models that address either users or charging stations, these approaches often overlook external factors such as weather and traffic conditions, which significantly impact driving patterns and energy consumption. The integration of these external variables into forecasting models is thus critical for enhancing prediction accuracy and infrastructure optimization. In this study, we conducted an in-depth analysis of EV charging patterns at highway rest areas. This study employed deep learning models, GRU, and LSTM architectures, trained using various data combinations. The experiments revealed that the inclusion of traffic data notably improves forecasting precision. In particular, the LSTM model demonstrated a 10.5% reduction in mean absolute percentage error(MAPE), decreasing the standard deviation from 4.95 to 3.87 when external factors were included.