Browse Articles

Elevated FT4 is associated with prevalent stroke in euthyroid adults: a cross‑sectional study

Resolving Abrikosov vortex entry in superconducting nanostring resonators via displacement-noise spectroscopy in cavity optomechanics

Edgeefficient human activity recognition using a quantized patchbased transformer

Abstract Human Activity Recognition (HAR) systems using deep learning have shown significant promise; however, deploying such models on edge computing systems remains challenging due to constraints in inference latency, memory footprints and computational capacity. This study proposes a lightweight, end-to-end patch-based Transformer architecture designed for efficient HAR for resource constrained edge environments, evaluated through an architectural ablation study and multiple quantization strategies using TensorFlow Lite. We conducted extensive experiments by varying patch lengths and Transformer encoder depths in order to evaluate the impact on classification accuracy, latency and model complexity. The results demonstrate that moderate temporal patching achieves an effective balance between temporal representation learning and computational efficiency. Experiments have been conducted on two widely open-source benchmark datasets, WISDM and PAMAP2, where raw sensor signals are pre-processed through segmentation, normalization, and overlapping windowing before being fed into a compact Transformer model with patch embedding and multi-head self-attention for feature extraction. The trained models are converted into FP32, FP16, INT8 dynamic range, and INT8 full formats to evaluate trade-offs between different metrics such as accuracy, model size, and inference latency. The baseline model achieves 96.77% accuracy on WISDM and 98.48% on PAMAP2, with consistently high macro F1-scores. Among all quantization variants, TFLite-INT8-Dynamic reduced model size by 95.55% and 89.80% for WISDM and PAMAP2, respectively and drastically improved inference latency by around 99% for both datasets with minimal accuracy degradation. These findings demonstrate that the proposed patch-based Transformer model for HAR has achieved an effective balance between recognition performance and computational efficiency, which indicates its strong deployment feasibility for resource constrained edge environments through TensorFlow Lite benchmarking and can provide a scalable solution for edge intelligence applications.

Convergent genomic trajectories shape adaptation to life on land across animal lineages

Construction and mechanism of histone deacetylation related gene prognosis model in esophageal carcinoma

Wavelength-encoded neuromorphic inference enabled by microcavity MoS2 photodetector arrays

Dual-mode deep reinforcement learning for safety-oriented MASS collision avoidance

Human imprints on global riverfronts

Trust as a human value and driver of organizational behavior in innovative companies across international rankings

Learning ordinality-aware multimodal representations for composite materials design

Visual fidelity–driven quality assessment of medical image translation

Abstract Automated and reliable image quality assessment (IQA) is essential for safe use of medical image synthesis in critical applications like adaptive radiotherapy, treatment planning, or missing-modality reconstruction, where unnoticed generative artifacts may adversely affect outcomes. We evaluated image-to-image translation quality by coupling large-scale visual quality assessment with explainable automated IQA modeling. Adversarial diffusion-based framework, SynDiff, was applied to four cross-modality synthesis tasks, including three inter-MR and a CBCT-to-CT translation. Using four-fold cross-validation, ten reference-based and eight no-reference IQA metrics were computed for all synthesized images. Visual IQA ratings were independently collected from thirteen raters using predetermined protocol and specialized image viewer enabling blinded, randomized six-point Likert scoring. Auto-Sklearn was employed to learn ensemble regression models mapping IQA metrics to visual consensus ratings, with separate models trained on reference-based and no-reference metrics. The models closely reproduced distribution and ordering of ratings, typically within ±0.5 Likert points. Reference-based models achieved higher agreement with visual ratings than no-reference models ( $$R^2$$ 0.75 vs. 0.59, resp.), although the latter remained unbiased and informative. Explainability analyses indicated that metrics quantifying structural similarity or fidelity (e.g., anatomical boundary preservation) and intensity-based contrast relationships between tissues were the strongest predictors. Overall, the results demonstrate that ensemble regression models can provide transparent, scalable, and clinically meaningful quality control for generative medical imaging.

Process-integrated engineered resting cells for biocatalytic production of rare natural sugars from sole methanol molecules

Machine learning-guided dose-time optimization and experimental validation enhance coumarin therapeutics in oncology

Immune correlates of risk for SARS-CoV-2 infection in children: a prospective, community-based cohort study

Chemotherapy-induced senescence promotes stroma stiffness and antioxidant adaptation to promote chemoresistance in pancreatic ductal adenocarcinoma

Abstract Chemoresistance in pancreatic ductal adenocarcinoma (PDAC) is partly driven by pathological stromal remodeling, yet the underlying mechanisms remain poorly understood. Here, we show that gemcitabine treatment induces tumor cell senescence and activates cancer-associated fibroblasts via the senescence-associated secretory phenotype, leading to progressive fibrotic matrix stiffening. This biomechanical reprogramming engages the mechanosensitive ion channel Piezo1, triggering metabolic rewiring that renders BRG1-positive tumor cells increasingly dependent on NRF2-mediated antioxidant defenses. Piezo1 signaling promotes NRF2 nuclear translocation and its chromatin-remodeling cooperation with BRG1, thereby upregulating SLC7A11-dependent antioxidant programs and suppressing ferroptosis. Notably, the combination of the senolytic agent ABT-263 with the ferroptosis inducer Erastin effectively dismantles BRG1–NRF2-driven gemcitabine resistance, alleviates stromal fibrosis, enhances T-cell infiltration, and suppresses tumor growth in vivo. This senolytic–ferroptosis approach exploits metabolic vulnerabilities in chemotherapy-aged PDAC and provides a mechanistic rationale for stroma-targeted combination therapies.

Scale-bridging interface design enables high-performance sustainable thermoelectrics

Enhanced nuclear fusion in the sub-keV energy regime

Abstract Nuclear fusion requires overcoming or traversing a repulsive Coulomb barrier of hundreds of kiloelectronvolts, rendering the probability of fusion at sub-keV energies vanishingly small. Yet in condensed matter, the electronic and structural environment of reacting nuclei can profoundly alter fusion rates. Here we demonstrate that deuterium–deuterium fusion within metallic foils exhibits a pronounced reaction yield plateau (i.e., a finite, non-vanishing yield floor) below 2 keV—in stark contrast to the expected exponential suppression at low energy. At the lowest energies measured, this corresponds to fusion yields enhanced by more than 10¹⁸ relative to bare-nucleus (unscreened) expectations. Using a dual-chamber platform that combines electrochemical deuterium loading with low-energy ion-beam bombardment, we observe this behavior in both palladium and titanium hydrides. These results reveal a previously unrecognized regime of low-energy nuclear reactions in solids, demonstrating that materials degrees of freedom can fundamentally renormalize tunneling probabilities and fusion cross-sections.

Coordination electrochemistry taming reversible hypervalent bromine redox for energetic six-electron-transfer lithium-bromine battery

Abstract Conversion-type static bromine batteries demonstrate promise in high output voltage and large capacity for rechargeable energy storage due to the inherent polyvalent reaction potential. Nevertheless, the combination of the limited two-electron 2Br - /Br 2 redox couple and the redox-inactive organic ligands presents a fundamental bottleneck for the overall specific energy. Herein, ethyl viologen dibromide is developed as an energetically active positive electrode for organic lithium-bromine batteries by efficient coordination chemistry, featuring an advanced six-electron redox mechanism triggered by both intercalation and conversion reactions. The activated redox couple of 2Br - /2Br + incubates a collaborative increase in capacity (632.8 mAh g −1 Br ) and discharge voltage (3.7 V). Besides, ethyl viologen undergoes reversible two-step intercalation and extraction of Li + ions, contributing to additional energy storage. Reciprocal spectroscopic characterizations and computational electrochemistry indicate the chemisorption effect and interhalogen confinement and detail the dynamic mass-charge transfer pathway. This work sets a paradigm worth emulating for designing high-performance halogen batteries.

A quantum-classical hybrid framework for optimal energy storage systems planning

Abstract The extensive deployment of power-electronics introduce spatial-temporal variability that can degrade voltage quality and operational reliability. Energy storage systems (ESS) can mitigate these effects through fast active and reactive power support, but their value is contingent on coordinated siting and sizing. Integrated formulations that minimize voltage deviations, reduce substation power-flow variability, and account for installation costs typically yield in large-scale mixed-integer optimization problems that are computationally burdensome for classical solvers and may yet not lead to the most optimum solution. To address these challenges, this paper proposes a two-stage hybrid quantum–classical planning framework that separates binary siting from continuous sizing and operation. In Stage I, the siting problem is reformulated as a Quadratic Unconstrained Binary Optimization model and solved via a hybrid quantum workflow. Acting as a “quantum sieve,” stochastic sampling generates a diverse set of candidate site combinations that classical single-point methods can overlook. In Stage II, selected site sets are evaluated using a classical convex solver (SOCP) to compute optimal ESS capacities and operating setpoints subject to network constraints, ensuring physical feasibility. Experiments on IonQ Forte hardware show grid-standard accuracy with industry-standard classical solvers. Although current hardware latencies limit performance in the NISQ era, the paper outlines scaling pathways and discusses key practical hurdles, including state-preparation overlap and higher-order cost couplings.

Federated graph learning with spatio-temporal dynamics for cross-border recommendation

Abstract Cross-border data sharing is strictly constrained by privacy regulations, which presents a critical challenge for recommendation systems due to the severe shortage of training data. Existing federated graph neural network methods predominantly rely on the federated averaging strategy, which struggles to handle the highly heterogeneous data encountered in scenarios characterized by user isolation and business homogeneity. To address this issue, this paper proposes FedSTAR, a privacy-preserving cross-border recommendation framework that integrates spatio-temporal dynamic modeling with federated graph neural networks. Its core innovations include the design of a dynamic sequential graph structure to capture the evolution of user preferences, the use of a multi-head attention mechanism to filter noisy neighbors in the spatial dimension, and the introduction of a personalized federated aggregation strategy to replace traditional FedAvg, thereby enabling adaptive fusion of heterogeneous multi-source data. Evaluations on three public datasets, Gowalla, Yelp 2018, and Amazon Book, demonstrate that FedSTAR achieves average improvements of 2–5 percentage points in Recall@20 and 1–3 percentage points in NDCG@20, respectively. Under the privacy constraint of exchanging only model updates, FedSTAR delivers both high accuracy and strong robustness, offering a secure and practically viable solution for cross-border recommendation.