Taeyoung Park

State-of-the-art graph-based action recognition frameworks, such as Spatio-Temporal Graph Convolutional Networks (ST-GCN) and Two-Stream Adaptive Graph Convolutional Networks (2s-AGCN), have advanced skeleton-based spatio-temporal modeling by learning dynamic joint-wise dependencies across frames. Nevertheless, when applied to the high-variability firefighter behavior analysis---where complex tool-body interactions and subtle intent-driven motions coexist---these models often struggle with fine-grained separability across large action classes. To address this, we propose Class-Aware Topology-Guided Graph Convolutional Network (CATS-GCN). The novelty of our approach lies in a principled framework by integrating (i) anatomical topology constraints, (ii) class-prototype guided attention, and (iii) multi-head sparsity-diversity regularization into a unified learning objective. We evaluated the proposed framework on a large-scale firefighter action dataset comprising 341 fundamental and drill-specific actions. Experimental results demonstrate that CATS-GCN outperforms GCN-based baselines, achieving the highest classification accuracy. Beyond recognition performance, owing to its modular structure and interpretable attention patterns, our framework offers practical utility for automated training systems by facilitating correct posture maintenance, injury risk reduction, and enhanced operational readiness.

Put options are known to be priced unusually high in the market, which we refer to as the overpriced put puzzle . This study proposes a quantum model (QM) that can explain such high put option prices as fair prices. Starting from a stochastic differential equation of stock returns, we convert the Fokker–Planck equation into the Schrödinger equation. To model the market force that always draws excess returns back to equilibrium, we specify a diffusion process corresponding to a QM with a delta potential. The results demonstrate that stock returns follow a Laplace distribution and exhibit power law in the tail. We then construct a closed-form solution for European put option pricing, determining that our model better explains the returns of the S&P 500 index and its corresponding put option prices than do geometric Brownian motion-based models. This study has significant implications for investors and risk managers, presenting a model that can potentially improve derivative pricing. Future studies can generalize the model assumptions by introducing asymmetric potential drawing back excess returns to equilibrium.

Channel identifiability (CID) refers to the ability to distinguish between individual channels in time series (TS) modeling. The absence of CID often results in producing identical outputs for identical inputs, disregarding channel-specific characteristics. In this paper, we highlight the importance of CID and propose Channel Normalization (CN), a simple yet effective normalization strategy that enhances CID by assigning distinct affine transformation parameters to each channel. We further extend CN in two ways: 1) Adaptive CN (ACN) dynamically adjusts parameters based on the input TS, improving adaptability in TS models, and 2) Prototypical CN (PCN) introduces a set of learnable prototypes instead of per-channel parameters, enabling applicability to datasets with unknown or varying number of channels and facilitating use in TS foundation models. We demonstrate the effectiveness of CN and its variants by applying them to various TS models, achieving significant performance gains for both non-CID and CID models. In addition, we analyze the success of our approach from an information theory perspective. Code is available at https://github.com/seunghan96/CN.

Background and aims: Amid questions about a perceived association between low‐to‐moderate alcohol consumption and reduced cardiovascular risk from recent studies, research examining the impact of alcohol consumption on cardiometabolic risk markers has reported inconsistent results. We measured the association between long‐term alcohol consumption and incident health risk conditions related to cardiometabolic risk markers.Design, setting and participants: A prospective cohort analysis was conducted using data from a population‐based cohort (2001–2020) of 4708 Koreans aged 40–69 years who were free of chronic diseases, including metabolic syndrome, at baseline.Measurements: Alcohol consumption and incident cases related to cardiometabolic risk markers were assessed biennially using interviewer‐administered questionnaires or health examinations. The average alcohol intake was calculated to reflect long‐term consumption. Alcohol consumption was categorized as none or rarely (0–2 g/day), light (2 < −16 g/day, within guideline limits), medium (16 < −32 g/day) and heavy (>32 g/day).Findings: After full adjustment for covariates, light, medium or heavy alcohol consumption was positively associated with metabolic syndrome [hazard ratio (HR) = 1.55, 95% confidence interval (CI) = 1.32–1.83], abdominal obesity (HR = 1.41, 95% CI = 1.17–1.70), hyperglycemia (HR = 1.91, 95% CI = 1.62–2.26), high blood pressure (HR = 2.04, 95% CI = 1.72–2.41) and hypertriglyceridemia (HR = 1.59, 95% CI = 1.30–1.93), with the results presented for heavy alcohol consumption only (p‐values for trend p‐values for interaction and p‐values for trend <0.05 for all cases).Conclusions: Long‐term light (within guideline limits), medium or heavy alcohol consumption appears to be positively and linearly associated with incident health risk conditions related to cardiometabolic risk markers. The threshold dose of alcohol for developing hyperglycemia and high blood pressure appears to be lower in women.

Background/Objectives: Studies examining the association between coffee consumption and glucose metabolism markers have reported inconsistent findings, and few have considered coffee type. We investigated the association between coffee consumption and glucose metabolism markers in Korean adults. Methods: A cross-sectional analysis was conducted using data from the Korea National Health and Nutrition Examination Survey (2019–2021), including 7453 adults aged 19–64 years. Coffee consumption was assessed using a 24 h dietary recall and categorized as black coffee or coffee with sugar and/or cream (non-drinkers, ≤1 cup/day, 2 cups/day, and ≥3 cups/day). Multivariable logistic regression models were used to assess the associations with glucose metabolism markers, including the homeostasis model assessment of insulin resistance (HOMA-IR). Results: The mean age of participants was 40.6 years (standard error: 0.19). After full adjustment for covariates, in women, consuming two or more cups of black coffee per day was inversely associated with elevated HOMA-IR (2 cups/day: OR = 0.73, 95% CI: 0.56–0.96; ≥3 cups/day: OR = 0.66, 95% CI: 0.44–0.99) and fasting insulin levels (2 cups/day: OR = 0.70, 95% CI: 0.54–0.91; ≥3 cups/day: OR = 0.64, 95% CI: 0.43–0.93), compared to no consumption, showing a significant linear trend (p-trend ≤0.01 for all cases). By coffee type, women who consumed more black coffee had lower odds of elevated HOMA-IR and fasting insulin levels (p-trend ≤0.02). In men or those consuming coffee with sugar and/or cream, no significant associations with glucose metabolism markers were observed. Conclusions: Our findings indicate that consuming two or more cups of black coffee per day is inversely associated with insulin resistance in Korean women.

Masked autoencoders (MAEs) have recently demonstrated effectiveness in tabular data imputation. However, due to the inherent heterogeneity of tabular data, the uniform random masking strategy commonly used in MAEs can disrupt the distribution of missingness, leading to suboptimal performance. To address this, we propose a proportional masking strategy for MAEs. Specifically, we first compute the statistics of missingness based on the observed proportions in the dataset, and then generate masks that align with these statistics, ensuring that the distribution of missingness is preserved after masking. Furthermore, we argue that simple MLP-based token mixing offers competitive or often superior performance compared to attention mechanisms while being more computationally efficient, especially in the tabular domain with the inherent heterogeneity. Experimental results validate the effectiveness of the proposed proportional masking strategy across various missing data patterns in tabular datasets.

Masked autoencoders (MAEs) have recently demonstrated effectiveness in tabular data imputation. However, due to the inherent heterogeneity of tabular data, the uniform random masking strategy commonly used in MAEs can disrupt the distribution of missingness, leading to suboptimal performance. To address this, we propose a proportional masking strategy for MAEs. Specifically, we first compute the statistics of missingness based on the observed proportions in the dataset, and then generate masks that align with these statistics, ensuring that the distribution of missingness is preserved after masking. Furthermore, we argue that simple MLP-based token mixing offers competitive or often superior performance compared to attention mechanisms while being more computationally efficient, especially in the tabular domain with the inherent heterogeneity. Experimental results validate the effectiveness of the proposed proportional masking strategy across various missing data patterns in tabular datasets. Code is available at: \url{https://github.com/normal-kim/PMAE}.

Seunghan Lee, Taeyoung Park. Korean J Appl Stat 2024;37:631-41. https://doi.org/10.5351/KJAS.2024.37.5.631