Unsupervised anomaly detection for earthquake detection on Korea high-speed trains using autoencoder-based deep learning models.

We propose a method for detecting earthquakes for high-speed trains based on unsupervised anomaly-detection techniques. In particular, we utilized autoencoder-based deep learning models for unsupervised learning using only normal training vibration data. Datasets were generated from South Korean high-speed train data, and seismic data were measured using seismometers nationwide. The proposed method is compared with the conventional Short Time Average over Long Time Average (STA/LTA) model, considering earthquake detection capabilities, focusing on a Peak Ground Acceleration (PGA) threshold of 0.07, a criterion for track derailment. The results show that the proposed model exhibit improved earthquake detection capabilities than STA/LTA for PGA of 0.07 or higher. Furthermore, the proposed model reduced false earthquake detections under normal operating conditions and accurately identified normal states. In contrast, the STA/LTA method demonstrated a high rate of false earthquake detection under normal operating conditions, underscoring its propensity for inaccurate detection in many instances. The proposed approach shows promising performance even in situations with limited seismic data and offers a viable solution for earthquake detection in regions with relatively few seismic events.

Synthetic Tabular Data Based on Generative Adversarial Networks in Health Care: Generation and Validation Using the Divide-and-Conquer Strategy.

BACKGROUND: Synthetic data generation (SDG) based on generative adversarial networks (GANs) is used in health care, but research on preserving data with logical relationships with synthetic tabular data (STD) remains challenging. Filtering methods for SDG can lead to the loss of important information. OBJECTIVE: This study proposed a divide-and-conquer (DC) method to generate STD based on the GAN algorithm, while preserving data with logical relationships. METHODS: The proposed method was evaluated on data from the Korea Association for Lung Cancer Registry (KALC-R) and 2 benchmark data sets (breast cancer and diabetes). The DC-based SDG strategy comprises 3 steps: (1) We used 2 different partitioning methods (the class-specific criterion distinguished between survival and death groups, while the Cramer V criterion identified the highest correlation between columns in the original data); (2) the entire data set was divided into a number of subsets, which were then used as input for the conditional tabular generative adversarial network and the copula generative adversarial network to generate synthetic data; and (3) the generated synthetic data were consolidated into a single entity. For validation, we compared DC-based SDG and conditional sampling (CS)-based SDG through the performances of machine learning models. In addition, we generated imbalanced and balanced synthetic data for each of the 3 data sets and compared their performance using 4 classifiers: decision tree (DT), random forest (RF), Extreme Gradient Boosting (XGBoost), and light gradient-boosting machine (LGBM) models. RESULTS: The synthetic data of the 3 diseases (non-small cell lung cancer [NSCLC], breast cancer, and diabetes) generated by our proposed model outperformed the 4 classifiers (DT, RF, XGBoost, and LGBM). The CS- versus DC-based model performances were compared using the mean area under the curve (SD) values: 74.87 (SD 0.77) versus 63.87 (SD 2.02) for NSCLC, 73.31 (SD 1.11) versus 67.96 (SD 2.15) for breast cancer, and 61.57 (SD 0.09) versus 60.08 (SD 0.17) for diabetes (DT); 85.61 (SD 0.29) versus 79.01 (SD 1.20) for NSCLC, 78.05 (SD 1.59) versus 73.48 (SD 4.73) for breast cancer, and 59.98 (SD 0.24) versus 58.55 (SD 0.17) for diabetes (RF); 85.20 (SD 0.82) versus 76.42 (SD 0.93) for NSCLC, 77.86 (SD 2.27) versus 68.32 (SD 2.37) for breast cancer, and 60.18 (SD 0.20) versus 58.98 (SD 0.29) for diabetes (XGBoost); and 85.14 (SD 0.77) versus 77.62 (SD 1.85) for NSCLC, 78.16 (SD 1.52) versus 70.02 (SD 2.17) for breast cancer, and 61.75 (SD 0.13) versus 61.12 (SD 0.23) for diabetes (LGBM). In addition, we found that balanced synthetic data performed better. CONCLUSIONS: This study is the first attempt to generate and validate STD based on a DC approach and shows improved performance using STD. The necessity for balanced SDG was also demonstrated.

Dancing on the inside: A qualitative study on online dance learning with teacher-AI cooperation.

Recent technologies have extended opportunities for online dance learning by overcoming the limitations of space and time. However, dance teachers report that student-teacher interaction is more likely to be challenging in a distant and asynchronous learning environment than in a conventional dance class, such as a dance studio. To address this issue, we introduce DancingInside, an online dance learning system that encourages a beginner to learn dance by providing timely and sufficient feedback based on Teacher-AI cooperation. The proposed system incorporates an AI-based tutor agent (AI tutor, in short) that uses a 2D pose estimation approach to quantitatively estimate the similarity between a learner's and teacher's performance. We conducted a two-week user study with 11 students and 4 teachers. Our qualitative study results highlight that the AI tutor in DancingInside could support the reflection on a learner's practice and help the performance improvement with multimodal feedback resources. The interview results also reveal that the human teacher's role is essential in complementing the AI feedback. We discuss our design and suggest potential implications for future AI-supported cooperative dance learning systems.

Real-time, in vivo skin cancer triage by laser-induced plasma spectroscopy combined with a deep learning-based diagnostic algorithm.

BACKGROUND: Although various skin cancer detection devices have been proposed, most of them are not used owing to their insufficient diagnostic accuracies. Laser-induced plasma spectroscopy (LIPS) can noninvasively extract biochemical information of skin lesions using an ultrashort pulsed laser. OBJECTIVE: To investigate the diagnostic accuracy and safety of real-time noninvasive in vivo skin cancer diagnostics utilizing nondiscrete molecular LIPS combined with a deep neural network (DNN)-based diagnostic algorithm. METHODS: In vivo LIPS spectra were acquired from 296 skin cancers (186 basal cell carcinomas, 96 squamous cell carcinomas, and 14 melanomas) and 316 benign lesions in a multisite clinical study. The diagnostic performance was validated using 10-fold cross-validations. RESULTS: The sensitivity and specificity for differentiating skin cancers from benign lesions using LIPS and the DNN-based algorithm were 94.6% (95% CI: 92.0%-97.2%) and 88.9% (95% CI: 85.5%-92.4%), respectively. No adverse events, including macroscopic or microscopic visible marks or pigmentation due to laser irradiation, were observed. LIMITATIONS: The diagnostic performance was evaluated using a limited data set. More extensive clinical studies are needed to validate these results. CONCLUSIONS: This LIPS system with a DNN-based diagnostic algorithm is a promising tool to distinguish skin cancers from benign lesions with high diagnostic accuracy in real clinical settings.

Correction: Using Wake-Up Tasks for Morning Behavior Change: Development and Usability Study.

[This corrects the article DOI: 10.2196/39497.].

Using Wake-Up Tasks for Morning Behavior Change: Development and Usability Study.

BACKGROUND: Early morning behaviors between waking up and beginning daily work can develop into productive habits. However, sleep inertia limits the level of human ability immediately after waking, lowering a person's motivation and available time for productive morning behavior. OBJECTIVE: This study explores a design for morning behavior change using a wake-up task, a simple assignment the user needs to finish before alarm dismissal. Specifically, we set two research objectives: (1) exploring key factors that relate to morning behavior performance, including the use of wake-up tasks in an alarm app and (2) understanding the general practice of affecting morning behavior change by implementing wake-up tasks. METHODS: We designed and implemented an apparatus that provides wake-up task alarms and facilities for squat exercises. We recruited 36 participants to perform squat exercises in the early morning using the wake-up tasks for 2 weeks. First, we conducted a generalized estimating equation (GEE) analysis for the first research objective. Next, we conducted a thematic analysis of the postsurvey answers to identify key themes about morning behavior change with the wake-up tasks for the second objective. RESULTS: The use of wake-up tasks was significantly associated with both the completion of the target behavior (math task: P=.005; picture task: P<.001) and the elapsed time (picture task: P=.08); the time to alarm dismissal was significantly related to the elapsed time to completion (P<.001). Moreover, the theory of planned behavior (TPB) variables, common factors for behavior change, were significant, but their magnitudes and directions differed slightly from the other domains. Furthermore, the survey results reveal how the participants used the wake-up tasks and why they were effective for morning behavior performance. CONCLUSIONS: The results reveal the effectiveness of wake-up tasks in accomplishing the target morning behavior and address key factors for morning behavior change, such as (1) waking up on time, (2) escaping from sleep inertia, and (3) quickly starting the desired target behavior.

The connection to the public's preferred sports analysis and physical education curriculum.

People have their favorite type of sport, but such preferences tend to be shared for nearly a lifetime. How this preference persists remains inconclusive; hence, this study attempts to determine why people have different viewpoints on sports. It is reasonable to infer that these differences arise from differences in culture, occupation, and race. Therefore, we collected the following data and conducted research in Korea, the United States, and Japan, countries with various differences. The types of sports that people play were collected through surveys and comparisons among sports networks. Namely, "Sport Classification," "The K-12 Physical Education System (textbooks)," "Survey (actual physical activity)," "Simple Notification Service (SNS) Activity" have been examined to deduce the reason why any particular sport is played. Firstly, Korea, the United States, and Japan conduct different physical education courses. Hence, the results affect people's preferences. Secondly, what people post on SNS and their actual physical activities are different. Thirdly, the degree of connection between sports-type varied as well. Lastly, sports that serve the purpose of being regarded as hubs among sports-type were common in Korea, the United States, and Japan.

What Makes Sports Fans Interactive? Identifying Factors Affecting Chat Interactions in Online Sports Viewing.

Sports fans are able to watch games from many locations using TV services while interacting with other fans online. In this paper, we identify the factors that affect sports viewers' online interactions. Using a large-scale dataset of more than 25 million chat messages from a popular social TV site for baseball, we extract various game-related factors, and investigate the relationships between these factors and fans' interactions using a series of multiple regression analyses. As a result, we identify several factors that are significantly related to viewer interactions. In addition, we determine that the influence of these factors varies according to the user group; i.e., active vs. less active users, and loyal vs. non-loyal users.