Natural Gradient Descent (NGD) has emerged as a promising optimization algorithm for training neural network-based solvers for partial differential equations (PDEs), such as Physics-Informed Neural Networks (PINNs). However, its practical use is often limited by the high computational cost of solving linear systems involving the Gramian matrix. While matrix-free NGD methods based on the conjugate gradient (CG) method avoid explicit matrix inversion, the ill-conditioning of the Gramian significantly slows the convergence of CG. In this work, we extend matrix-free NGD to broader classes of problems than previously considered and propose the use of Randomized Numerical Linear Algebra (RandNLA) techniques for efficient preconditioning of the inner CG solver. The resulting algorithms demonstrate substantial performance improvements over existing NGD-based methods on a range of PDE problems discretized using neural networks, and offer competitive results compared to other state-of-the-art optimizers.
This study examined the psychometric properties and responsiveness to training of a comprehensive Advanced Theory of Mind (AToM) Battery in middle childhood. In Study 1, 168 eight- to nine-year-old children participated (Mage = 8.61 years, SD = 0.64; 88 girls, 79 boys, one preferred not to say). Children completed the AToM Battery, the Strange Stories task, and measures of language, inhibition, working memory, and nonverbal intelligence. Confirmatory factor analyses (CFA) largely supported the proposed three-factor structure. At the same time, the CFA indicated that several items, particularly within the Social Reasoning component, showed comparatively weak loadings. The AToM Battery showed moderate convergent validity with the Strange Stories task, r = .37, p < .001. Language ability significantly predicted overall AToM performance, R2 = .16, with effects primarily driven by the Social Reasoning component, R2 = .15. In Study 2, 126 children completed the battery at pretest, posttest, and follow-up after participating in either a conversation-based training or a control training. Linear mixed-effects models indicated general improvements over time across groups. Training-specific gains emerged most consistently for Recognition of Transgressions of Social Norms at posttest [t(247.20) = 3.45, p < .001] and follow-up [t(247.20) = 3.23, p = .001]. Stability correlations in the control group indicated moderate temporal stability over a three-month interval, r = .55, p < .001. Overall, the findings support the AToM Battery as a multidimensional and psychometrically useful measure of advanced social-cognitive reasoning, while highlighting the importance of considering language demands and subscale-specific responsiveness.
Growing evidences reporting increasing diffusion and toxic effects of synthetic cathinones (SCs) within the recreational market of novel psychoactive substance (NPS) upstretched public health concerns. However, However, the pathophysiological responses underlying their toxicity remain poorly understood. In particular, despite its increasing circulation, data regarding 3-Chloromethcathinone (3-CMC) toxicological profile and pharmacological properties remains limited. To fill this gap, this in vivo study explored systemic effects of 3- CMC repeated exposures in mice, comparing sex differences. Male and female mice received i.p. injections of 3-CMC once weekly for five consecutive weeks (total of 35 days), with doses escalating from 0.1 to 1, 6, 10 and 30 mg center dot kg-1 in successive weeks (vehicle: 0.004 mL center dot g-1, i.p.), for a total of five injections (D1, D2, D3, D4 and D5) with a 6-day period of washout between each consecutive injection, and were then assigned to blood/tissue analyses. Blood and urine biochemical parameters were measured, paralleled by histopathological and immunohistochemical analyses on different target organs, i.e. liver, kidney and spleen. Our data demonstrated significant alterations in blood and urine biochemical parameters, accompanied by histopathological changes in all evaluated tissues, both in male and female animals, paralleled by enhanced cell death pathways activation. From a translational perspective, these preclinical findings provide crucial evidences concerning 3-CMC systemic toxicity, developing knowledge on potential health risks posed by this NPS. These data may also lead the way for further researches on CMC compounds and related toxidromes, aiding development of effective clinical strategies helping physicians in patients' management after exposure to SCs.
Dengue, an Aedes mosquito-borne viral infection, is on the rise with climate and demographic change. In 2023, WHO declared dengue its highest grade of emergency, following the largest number of cases and deaths in recorded history. This emergency highlighted the absence of safe and effective dengue-specific treatments as a gap in the product landscape for care for people with dengue. To accelerate development of dengue-specific treatments for wide implementation and access in all dengue-endemic countries, WHO conducted a landscape analysis of the dengue therapeutics pipeline and convened a global expert and public consultation to develop target product profiles for treatments for non-severe and severe dengue. These target product profiles provide strategic guidance for product developers, regulators, procurement agencies, and funders on the intended use, target populations, and key product characteristics of dengue-specific treatments required to meet public health needs, setting clear targets to drive development of dengue drugs accessible to those who need them the most.
Muscular fatigue can significantly impair joint stability, alter proprioception, and increase injury risk in multidirectional sports such as football. While superficial electromyography (sEMG) is widely used for fatigue detection, signal non-stationarity during dynamic movements limits the reliability of traditional methods. This study assessed the potential of complexity-based sEMG features in capturing myoelectric manifestations of fatigue during a dynamic task incorporating sport-related movement patterns, while examining variability in lower-limb joint kinematics. Forty-five football players were monitored using a synchronised wearable system comprising sEMG and Inertial Measurement Units (IMUs) during a running activity with unpredictable directional changes. Spectral sEMG features (Median Frequency and Instantaneous Median Frequency) were compared with complexity-based indices (Sample Entropy, SE; Permutation Entropy) across six muscles. Results suggest that SE may offer enhanced robustness against motion-related non-stationarity, exhibiting a progressive decrease across muscles. Specifically, the most consistent fatigue-related adaptations were observed in the soleus and the vastus medialis (p < 0.001). Concurrently, alterations in joint motion variability patterns were observed in the hip, knee, and ankle during the running cycle, which may reflect fatigue-induced adaptations. These findings suggest that SE is a promising indicator for monitoring neuromuscular adjustments in dynamic conditions. Furthermore, exploring the co-occurrence between complexity-based sEMG trends and IMU-derived kinematics offers a valuable approach for contextualising adaptations in neuromuscular control during a challenging dynamic task. Algorithmic estimates suggest the theoretical feasibility of near-real-time computation, though full system validation in real-world sports environments remains a prerequisite for formally recommending coaching applications or injury-prevention strategies.