Kuan Tao

Spiking Neural Networks (SNNs) are promising energy-efficient models and powerful framworks of modeling neuron dynamics. However, existing binary spiking neurons exhibit limited biological plausibilities and low information capacity. Recently developed ternary spiking neuron possesses higher consistency with biological principles (i.e. excitation-inhibition balance mechanism). Despite of this, the ternary spiking neuron suffers from defects including iterative information loss, temporal gradient vanishing and irregular distributions of membrane potentials. To address these issues, we propose Complemented Ternary Spiking Neuron (CTSN), a novel ternary spiking neuron model that incorporates an learnable complemental term to store information from historical inputs. CTSN effectively improves the deficiencies of ternary spiking neuron, while the embedded learnable factors enable CTSN to adaptively adjust neuron dynamics, providing strong neural heterogeneity. Furthermore, based on the temporal evolution features of ternary spiking neurons' membrane potential distributions, we propose the Temporal Membrane Potential Regularization (TMPR) training method. TMPR introduces time-varying regularization strategy utilizing membrane potentials, furhter enhancing the training process by creating extra backpropagation paths. We validate our methods through extensive experiments on various datasets, demonstrating remarkable performance advances.

This study investigated differences in energy-expenditure (EE) modeling between badminton players of varying competitive levels during aerobic training. It evaluated the impact of sensor quantity and sample size on prediction model accuracy and generalizability, providing evidence for personalized training-load monitoring. Fifty badminton players (25 elite, 25 enthusiasts) performed treadmill running, cycling, rope skipping, and stair walking. Data were collected using accelerometers (waist, wrists, ankles), a heart rate monitor, and indirect calorimetry (criterion EE). Multiple machine learning models (Linear Regression, Bayesian Ridge Regression, Random Forest, Gradient Boosting) were employed to develop EE prediction models. Performance was assessed using R², mean absolute percentage error (MAPE), and root mean square error (RMSE), with further evaluation via the Triple-E framework (Effectiveness, Efficiency, Extension). Elite athletes demonstrated stable, coordinated movement patterns, achieving the best values for R² and the smallest errors using minimal core sensors (typically dominant side). Enthusiasts required multi-site sensors to compensate for greater execution variability. Increasing sensors beyond three yielded no performance gains; optimal configurations involved 2–3 core accelerometers combined with heart rate data. Expanding sample size significantly enhanced model stability and generalizability (e.g., running task R² increased from 0.49 (N = 20) to 0.95 (N = 40)). Triple-E evaluation indicated that strategic sensor minimization coupled with sufficient sample size maximized predictive performance while reducing computational cost and deployment burden. Competitive level significantly influences EE modeling requirements. Elite athletes are suited to a “low-sensor, small-sample” scenario, whereas enthusiasts necessitate a “multi-sensor, large-sample” strategy.

Objective: Dance, as a globally practiced physical activity, presents challenges in accurately assessing energy expenditure due to its diverse styles and tempos. Traditional methods, relying on empirical formulas within ActiGraph accelerometers, often result in significant biases. While multiple wearable sensors have been introduced to mitigate these biases, they increase model complexity. Methods: This study proposes the Triple-E principle—Effectiveness, Efficiency, and Extension—as a framework for developing state-of-the-art (SOTA) machine learning models aimed at accurately estimating energy expenditure, while minimizing model complexity and optimizing sensor placement. To validate the proposed approach, we recruited a cohort of 250 participants (mean age: 63.0 ± 6.0 years), each performing ballroom, aerobic, or square dance routines. Participants were fitted with ActiGraph wGT3X-BT accelerometers at five anatomical locations, along with the CORTEX MetaMax 3B gas analyzer for metabolic data collection. We analyzed 311 physiological signal sequences and 1,555 acceleration count sequences. Results: Empirical formulas were proved inaccurate for dance energy expenditure, with Mean Absolute Percentage Error (MAPE) exceeding 50% and Root Mean Squared Error (RMSE) surpassing 3.23. A bidirectional stepwise regression model incorporating heart rate or triaxial motion sequences from accelerometers achieved an average goodness-of-fit of 0.73, identifying optimal accelerometer sites based on Efficiency principle. A random forest regression model minimized errors to 5% (MAPE) and 0.33 (RMSE) with data from all sites. Notably, wrist accelerometers and heart rate alone provided sufficient accuracy (RMSE: 0.35-0.36), highlighting a trade-off between Effectiveness and Efficiency. A deep-learning network pipeline based on the Extension principle automatically extracted features, achieving an average RMSE to 0.15. Conclusion: This study introduces a pioneering quantitative and unified model assessment system. Thoroughly analyzed and validated in the context of dance, the research offers detailed explanations of the most effective, efficient, and extensive models.

Background: The mass and strength of skeletal muscle decline with age, leading to its progressive dysfunction. High-throughput metabolite profiling provides the opportunity to reveal metabolic mechanisms and the identification of biomarkers. However, the role of amino acid metabolism in possible sarcopenia remains unclear. Objectives: The aim of this study included exploring variations in plasma amino acid concentrations in elderly individuals who have possible sarcopenia and further attempting to characterize a distinctive plasma amino acid profile through targeted metabolomics. Methods: A cross-sectional, correlational research design was used for this study. Thirty possible-sarcopenic elderly participants were recruited (n = 30), as determined by the Asian Working Group for Sarcopenia (AWGS). Meanwhile, a reference group of non-sarcopenic (sex-, age-, and Appendicular Skeletal muscle Mass Index (ASMI)-matched non-sarcopenic controls, n = 36) individuals was included to compare the potential differences in metabolic fingerprint of the plasma amino acids associated with sarcopenia. Both groups were conducted the body composition analysis, physical function examination, and plasma amino acid-targeted metabolomics. The amino acids in plasma were measured using ultra-performance liquid chromatography-tandem mass spectrometry (UPLC-MS–MS). Also, orthogonal partial least-squares-discriminant analysis (OPLS-DA) was applied to characterize the plasma amino acid profile. Results: With respect to Handgrip Strength (HGS), the Five-Repetition Chair Stand Test (CS-5), the Six-Minute Walking Test (6MWT), the arm curl, the 30 s-Chair Stand Test (CST), the 2-Minute Step Test (2MST), the Timed Up-and-Go Test (TUGT), there was a decline in skeletal muscle function in the possible-sarcopenic group compared to the non-sarcopenic group. The mean plasma concentrations of arginine, asparagine, phenylalanine, serine, lysine, glutamine, and threonine were significantly lower in the possible sarcopenia group, whereas cirulline, proline, serine, and glutamic acid concentrations were higher. According to the multi-analysis, glutamine, serine, lysine, threonine, and proline were determined as the potential markers that indicated possible sarcopenia. Conclusions: The findings characterize significantly altered plasma amino acid metabolisms in the elderly with possible sarcopenia, which aids to screening people who are at a high risk of developing condition, and motivating to design new preventive and therapeutic approaches.

The Perception Neuron Studio (PNS) is a cost-effective and widely used inertial motion capture system. However, a comprehensive analysis of its upper-body motion capture accuracy is still lacking, before it is being applied to biomechanical research. Therefore, this study first evaluated the validity and reliability of this system in upper-body capturing and then quantified the system’s accuracy for different task complexities and movement speeds. Seven participants performed simple (eight single-DOF upper-body movements) and complex tasks (lifting a 2.5 kg box over the shoulder) at fast and slow speeds with the PNS and OptiTrack (gold-standard optical system) collecting kinematics data simultaneously. Statistical metrics such as CMC, RMSE, Pearson’s r, R², and Bland–Altman analysis were utilized to assess the similarity between the two systems. Test–retest reliability included intra- and intersession relations, which were assessed by the intraclass correlation coefficient (ICC) as well as CMC. All upper-body kinematics were highly consistent between the two systems, with CMC values 0.73–0.99, RMSE 1.9–12.5°, Pearson’s r 0.84–0.99, R² 0.75–0.99, and Bland–Altman analysis demonstrating a bias of 0.2–27.8° as well as all the points within 95% limits of agreement (LOA). The relative reliability of intra- and intersessions was good to excellent (i.e., ICC and CMC were 0.77–0.99 and 0.75–0.98, respectively). The paired t-test revealed that faster speeds resulted in greater bias, while more complex tasks led to lower consistencies. Our results showed that the PNS could provide accurate enough upper-body kinematics for further biomechanical performance analysis.