搜索结果：Work

The physical scaling of photonic matrix-vector multiplication hardware for deep neural network acceleration is fundamentally limited by accumulated optical losses, crosstalk noise, and the prohibitive footprint of conventional devices such as Mach-Zehnder interferometers. Here we present LightPro, a fully programmable linear photonic processor designed to optimize scalability, power efficiency, and area footprint. At its core, our architecture integrates a neural architecture search and pruning framework with tunable phase-change material directional couplers. By thermally modulating the phase-change material state, we dynamically adjust coupling coefficients to achieve precise splitting ratios, facilitating highly optimized topologies for matrix-vector multiplication operations. The underlying phase-change material-based devices are evaluated using numerical multiphysics simulations and compact models, which are validated against reported experimental data from prior work. System-level evaluations demonstrate that the neural architecture search-optimized LightPro architectures achieve up to an 85% footprint reduction and a greater than 50% decrease in power consumption. Network scaling evaluations using handwritten digit and Gaussian datasets yield an inference accuracy degradation of less than 5%. Experimental prototyping on a commercial photonic processor validates the computational accuracy of LightPro, establishing a scalable and efficient pathway for next-generation photonic artificial intelligence accelerators.

This study proposes a data-driven experimental decision framework to identify the most suitable sustainable supplementary materials for green concrete, aiming to reduce cement usage, industrial waste burden, and environmental impacts in the construction sector. It experimentally evaluates compressive strength, split tensile strength, flexural strength, and ultrasonic pulse velocity (UPV) of green concrete incorporating waste materials including silica fume, GGBS, metakaolin, granite dust, rice husk ash, ceramic waste, marble powder, coconut shell powder, plastic waste, and bottom ash. A hybrid methodology integrating Pearson correlation, Analytical Hierarchy Process (AHP), and k-means clustering was developed to capture complex interrelationships. Correlation-based dependency analysis was incorporated into AHP to generate objective performance weightages, where compressive strength was ranked highest (37%), followed by flexural strength (25%), UPV (22%), and split tensile strength (16%). K-means clustering then categorized materials into best and worst performance groups. The findings revealed silica fume as the most optimal and balanced material, achieving 48.5&#xa0;MPa compressive strength, 4.0&#xa0;MPa split tensile strength, 7.5&#xa0;MPa flexural strength, and 4400&#xa0;m/s UPV, indicating superior structural performance and durability potential. ANOVA confirmed strong statistical distinction between clusters (p&#x2009;<&#x2009;0.0001), validating the robustness of the classification. The main contribution of this work lies in introducing a scalable machine-learning-assisted multi-criteria framework that objectively ranks sustainable cement replacement materials, enabling reliable selection for high-performance green concrete design.

In this paper, a distributed robust adaptive confined fault-tolerant optimal control method based on deep neural networks is proposed, aiming to solve the complexity and uncertainty problems in human height and weight prediction. Note that the term 'control' in this work refers to feedback regulation of the iterative learning/optimization dynamics of the predictor (iteration domain), rather than controlling the physical time evolution of human height/weight. In the field of public security technology, accurate prediction of individual physiological characteristics has important application value, especially in crime prevention, individual identification, and behavior analysis. Traditional prediction methods often perform erratically in the face of data noise, environmental changes, and outliers. To this end, this paper combines deep learning and fault-tolerant control theory to propose an efficient and reliable prediction framework by optimizing the robustness and adaptive ability of the predictor. By introducing a limited fault-tolerant mechanism, it can maintain high prediction accuracy and stability under various perturbations and incomplete data conditions. Moreover, we evaluate the proposed framework from three complementary dimensions-statistical similarity, overall predictive performance, and minority-class detection ability-and explicitly acknowledge that these criteria may exhibit trade-offs: improved distributional similarity does not necessarily translate into better decision boundaries, and optimizing overall performance can conflict with minority detection (e.g., recall/F1). Simulation and experimental results show that after 2000 rounds of iterative optimization, the normal and fault-tolerant prediction accuracies of human height for finger length of left and right hands are 98.4% and 97.7%, respectively, and the normal and fault-tolerant prediction accuracies of human body weight are 98.2% and 97.5%, respectively, by combining the 372 sets of data with 30% of data loss caused by human. The accuracy of normal and fault-tolerant prediction of human height was 90.8% and 89.2% for the finger length of the left hand, and the accuracy of normal and fault-tolerant prediction of human weight was 85.6% and 83.3%, respectively. The normal and fault-tolerant prediction accuracies of human height for the finger length of the right hand were 96% and 95.3%, and the normal and fault-tolerant prediction accuracies of human weight were 94.4% and 93.5%, respectively. These findings are most directly applicable to small-to-medium tabular datasets with moderate class imbalance and limited minority samples, which matches the regimes evaluated in this study. This study provides a new idea and technical path for biometric prediction and analysis in the field of public security technology, which has important theoretical significance and practical value.

Papillary thyroid carcinoma (PTC), though genetically characterized, lacks comprehensive epigenomic profiling. Here we present the first combined analysis of 5-methylcytosine (5mC), 5-hydroxymethylcytosine (5hmC), and 6-methyladenine (6mA) in PTC pathogenesis. Using mDIP-seq on matched tumor and adjacent normal tissues from six treatment-na&#xef;ve patients, we mapped genome-wide distributions of these modifications, revealing distinct spatial patterns and tumor-specific differentially methylated regions (DMRs) with modification-specific functional partitioning. Silhouette-guided feature selection refined 7,117 DMRs to a 187-locus diagnostic panel achieving robust tumor-normal discrimination, with chromosomes 16 and 19 as recurrent epigenetic hotspots (2.1-7.9-fold enrichment). Integration of TCGA data revealed mutation-dependent epigenetic-transcriptional associations: BRAF-mutant tumors showed selective upregulation of epigenetically dysregulated genes, whereas RAS-mutant tumors displayed distinct transcriptional patterns. Co-expression network analysis highlighted disrupted transcriptional coordination in tumors. This work establishes a multidimensional epigenetic framework for PTC with combinatorial diagnostic signatures and condition-specific associations between DNA modifications and driver mutations (BRAF vs. RAS).

Variational autoencoders (VAEs) combined with neural ordinary differential equations provide a flexible framework for exploring neural latent-variable models in population pharmacokinetics. In this work, we investigate an empirical Bayes VAE formulation that integrates encoder-decoder architectures with covariate-dependent population priors, enabling correlated latent representations and probabilistic inference. We evaluate the proposed framework using controlled simulation studies and a small clinical benchmark dataset. The simulation experiments assess the ability to recover known population structures and covariate effects, while the clinical study evaluates subject-specific prediction and model diagnostics. In simulation studies with correlated individual parameters, the empirical Bayes VAE consistently captured population-level variability, whereas a fixed-prior VAE baseline exhibited systematic biases. In our experiments, extrapolation beyond the training dosing schedules showed more stable predictive behavior when using the proposed input-response normalization, relative to models trained without normalization, within a limited range. Diagnostic analyses indicated clear relationships between inferred latent variables and true parameters, and estimated observation noise was consistent with simulated values. In the clinical case study, cross-validation experiments suggested predictive performance comparable to previously reported neural ODE-based approaches. Overall, the results illustrate the feasibility of combining empirical Bayes inference with neural ODE-based decoders for population modeling. The proposed framework should be viewed as a methodological proof-of-concept, highlighting both the potential and the current limitations of variational neural approaches in pharmacometric applications.

Large-scale artificial intelligence (AI) models achieve notable performance in computer vision but require substantial computational resources, limiting their deployment on edge devices1,2. Optical neural networks (ONNs) promise reduced latency and energy consumption by making use of the inherent parallelism of light3. However, present ONNs struggle to scale and are confined to simple tasks, owing to the challenges of replicating exact algebraic operations of digital models using physical (analogue) systems. This work introduces a new paradigm that directly embeds core computer vision principles, including similarity-based recognition, attention-guided perception and detail-context fusion, into a large-scale optical metasurface. By unifying optical physics with these computer vision fundamentals, we develop a photonic-electronic engine that overcomes scalability and generality barriers, enabling high-accuracy, general-purpose computer vision at the edge. The resulting system combines a 41-million-parameter optical metasurface front end with a co-designed, ultraefficient 87,000-parameter digital back end, outperforming many digital models with tens of millions of parameters across object detection, segmentation, 3D reconstruction and video understanding. We build a deployable prototype and demonstrate real-time edge visual processing in natural scenes. This work represents a path towards practical optical computing for general vision tasks in complex natural environments, enabling a new paradigm for low-energy, low-latency, real-time on-device vision intelligence.

A flexible, lightweight and low-cost enzymatic bioanode was developed using a screen-printed silver conductive transparency sheet for enzymatic biofuel cell (EBFC) applications. In this work, indole was electrochemically polymerized directly onto the conductive substrate to form a polyindole (PIn) matrix capable of simultaneously entrapping glucose oxidase (GOx) and redox mediator vitamin K3 (VK3). The study integrates a disposable transparency-sheet platform with an electroactive PIn network that promotes efficient enzyme immobilization and enhanced electron transfer for glucose bioelectrocatalysis. The synergistic interaction between PIn, VK3 and GOx significantly improved charge-transfer kinetics and stabilized the bioelectrocatalytic interface, resulting in enhanced electrochemical performance. The fabricated PIn/VK3/GOx bioanode exhibited a current density of 1.18&#xa0;mA cm-&#x2009;2 in 40 mM glucose solution, demonstrating efficient glucose-dependent electrocatalytic activity. The conductive PIn framework facilitated rapid electron transport between the buried active sites of GOx and the electrode surface, while VK3 acted as an efficient and biocompatible electron shuttle. The bioanode also displayed good electrical stability, semiconducting behavior, and favorable electrochemical characteristics, highlighting its suitability for flexible and wearable bioelectronic applications.

The large-scale use of nickel (Ni [II]) in many industrial processes results in the production of Ni-containing wastes, posing a hazardous impact on public health and the environment. The current study highlights Aspergillus flavus's potential for environmental bioremediation and the impact of Ni stress on the growth kinetics, removal efficiency and bioaccumulation. The results showed that the biomass of A. flavus decreased at all tested Ni (II) concentrations as compared to the control. Additionally, the minimal inhibitory concentration (MIC) of A. flavus was 4500 mg L-&#x2009;1 on solid medium. The Ni (II) removal rate of A. flavus increased with increasing Ni (II) levels and peaked at 63.76% at 750 mg L-&#x2009;1, while the maximum bioaccumulation capacity (Q) reached 83.07 mg g-&#x2009;1 at 1750 mg L-&#x2009;1. Moreover, Ni (II) stress caused an increase in malondialdehyde (MDA) (137.74% at 1750 mg L-&#x2009;1). Antioxidant enzyme activities and secondary metabolites significantly increased to counteract oxidative stress before declining at extreme concentrations. Also, Fourier-transform infrared spectroscopy (FTIR) demonstrated that amide, phosphate, and sulfur groups participate in Ni (II) biosorption as accumulation sites. Furthermore, energy-dispersive X-ray (EDX) spectroscopy confirmed the presence of Ni (II) on the mycelial surface, with a measured weight% of 9.44%. Notably, our study reveals a synergistic network of defense mechanisms that support A. flavus's resilience. These results emphasize the work's novelty and substantiate the isolate's use as a feasible option for long-term bioremediation.

The persistence of biological material on various substrates over extended periods of time and the visualization of cellular degradation are an important area of consideration in DNA-TPPR research. This study investigates long-term cell persistence of touch deposits and the ability to visualise them on different substrates. A key focus of the study was optimisation of the Diamond Nucleic Acid Dye&#x2122; (DD) application to minimise disruption to cells while ensuring effective cell visualisation. Three spray methods were tested across seven distances, with performance evaluated based on fluorescence intensity and the spread of cellular material outside of the deposit circle.Using the optimised method persistence of touch cells on six substrates: glass, plastic, melamine, aluminium, leather, and cotton were assessed at several timepoints up to a year including effect of respraying.Results revealed diminishing and significant variation in. cell persistence across substrates, with cotton and leather displaying lowest persistence. Further, the results of the re-spraying, point to the need for respraying items after initial staining to ensure the best visualisation of cellular material over time. The DNA amounts in the deposits were then assessed, showing that the amount of DNA recovered was considerably less than what would be expected based on the cell counts.This research provides valuable baseline data for forensic caseworkers to prioritise, where possible, sample collection based on substrate-specific cell persistence. Additionally, the study aids the work towards establishing best practices for Diamond Dye&#x2122; application to maximise visualisation efficiency with minimal cell disruption.

Electron Wigner solids (WSs)1-12 provide an ideal system for understanding the competing effects of electron-electron and electron-disorder interactions, a central unsolved problem in condensed matter physics. Progress in this topic has been limited by a lack of single-defect-resolved experimental measurements as well as accurate theoretical tools to enable realistic experiment/theory comparison. Here we overcome these limitations by combining atomically resolved scanning tunnelling microscopy (STM) with neural-quantum-state quantum Monte Carlo (NQS-QMC) simulation of disordered 2D electron WSs to discover new disorder-induced physical regimes of correlated electron behaviour. STM was used to image the electron density (ne)-dependent evolution of electron WSs in gate-tunable bilayer MoSe2 (BL-MoSe2) devices with varying long-range (nLR) and short-range (nSR) disorder densities. These images were compared with NQS-QMC simulations using realistic disorder maps extracted from experiment, thus allowing the roles of different disorder types to be disentangled. We identify two distinct physical regimes for disordered electron WSs that depend on nSR. For nSR&#x2009;&#x2272;&#x2009;ne, the WS behaviour is dominated by long-range disorder and features extensive mixed solid-liquid phases, a new type of local re-entrant melting/crystallization and prominent Friedel oscillations. By contrast, when nSR&#x2009;&#x226b;&#x2009;ne, these features are suppressed and a more robust amorphous WS phase emerges that persists to higher ne, highlighting the importance of short-range disorder in this regime. Our work establishes a powerful framework for studying disordered quantum solids through a combined experimental-theoretical approach.

Serine/threonine kinases of the Hanks family are key regulators of bacterial physiology. Among them, membrane-associated PASTA-Hanks kinases govern bacterial cytokinesis and morphogenesis, yet their activation mechanism remains unclear. Here, we report crystal structures of the catalytic domain of the PASTA-Hanks kinase StkP of the human pathogen Streptococcus pneumoniae, carrying phosphoablative or phosphomimetic mutations in its activation loop. These structures demonstrate that phosphorylation of two threonine residues modulates the activation loop's organization and dynamics and reveal an alternative mode of dimerization of the catalytic domain. Analytical ultracentrifugation, SAXS and cell imaging allow to propose a model postulating that the local concentration of StkP at the division septum promotes an inactive dimeric state in which the activation loop hampers substrate binding. The reorganization into active dimers would activate StkP and allow endogenous substrate phosphorylation. This work thus provides a mechanistic framework of the regulation of PASTA Hanks kinase for the regulation of bacterial cell division.

Gravitational water vortex turbines (GWVTs) have emerged as promising hydrokinetic technology for energy extraction in riverine systems characterized by low flow velocity and ultra-low head, where conventional hydropower solutions are not feasible. In this study, a three-dimensional computational fluid dynamics (CFD) model is developed to investigate turbine performance within a conical basin configuration. The numerical framework, validated against published reference data, is used to systematically evaluate the influence of airfoil axial spacing, chord sizing, and airfoil profile, on vortex structure and power coefficient. The results demonstrate that basin-induced flow control plays a critical role in enhancing tangential momentum transfer to the rotor. Among the investigated cases, a configuration employing a NACA0024 ducted airfoil with a chord size of 35&#xa0;mm and an axial spacing of 50&#xa0;mm yielded the highest power coefficient of 0.419 compared to the baseline basin power coefficient of 0.313 and reached a maximum value of 0.736 at higher inlet velocities. Performance analysis based on tip speed ratio&#xa0;(TSR) further identified an optimal low-TSR operating range consistent with gravitational water vortex turbine characteristics. Compared with previous GWVT studies that focus on turbine geometry, the present work highlights the effectiveness of basin wall airfoil design as an impactful strategy for performance enhancement in low-head hydrokinetic applications.

Distinct mitophagy pathways can eliminate not only damaged mitochondria but also healthy ones. In Mitochondrial DNA Depletion Syndrome 13 (MTDPS13), dysregulated BNIP3/NIX-driven mitophagy of functional mitochondria is thought to be the key pathological driver. Patient mutations in the E3 ubiquitin ligase FBXL4 impair the proteasomal degradation of the mitophagy receptors BNIP3 and NIX, causing their accumulation and excessive mitophagy. As a result, mitochondrial content and oxidative phosphorylation decline sharply across multiple tissues, leading to early mortality, with no effective treatments currently existing. Here, we build on our work showing that AMPK can inhibit mitophagy via sequestration of the ULK1 autophagy-initiating kinase ULK1 and demonstrate that it is also critically relevant for mitophagy induced by FBXL4 disruption. Using FBXL4-deficient cells, as well as fibroblasts derived from MTDPS13 patients and a chemically-induced mouse model, we show that small molecule AMPK activation inhibits BNIP3/NIX-mediated mitophagy and recovers functional mitochondrial content. This work therefore validates AMPK as a realistic target in treating MTDPS13.

In the era of smart agriculture, agricultural information and related event news spread rapidly across social media and online platforms. Efficient classification of such massive agricultural texts holds practical value for optimizing production decisions and enhancing risk early warning. The wide dissemination of these texts also raises the demand for accurate and timely content analysis. However, the domain vocabulary sparsity, semantic ambiguity, and annotated data scarcity of agricultural texts pose notable challenges for classification tasks. Existing zero-shot classification methods based on large language models predominantly design prompt templates from a single perspective, making it difficult to simultaneously cover multiple levels of text analysis, while classification judgments also lack cross-validation mechanisms from multi-dimensional information. To address these issues, this paper proposes a multi-perspective prompt fusion framework that constructs multiple complementary analytical perspectives based on hierarchical differences in text analysis, forming multi-level coverage from local terminology features to overall thematic attribution, and guiding large language models to perform classification judgment from different cognitive dimensions. This paper further constructs a dual-dimensional consistency measurement mechanism integrating decision-level and confidence-level metrics, and employs an adaptive threshold-driven disagreement detection and arbitration mechanism to conduct secondary judgment on low-consistency samples. Experiments on the public datasets PestObserver-France and HumAID demonstrate that the proposed method achieves classification accuracy of 85.3% and 68.5%, corresponding to improvements of 3.2% and 3.7% over the strongest baseline ChatAgri. The expected calibration error decreases from 0.135 and 0.148 to 0.098 and 0.112, representing reductions of 27% and 24% respectively. Perspective heterogeneity analysis shows that the pairwise prediction agreement rate among the five perspectives is 18 to 21 percentage points lower than that of the five-path sampling of Self-Consistency, confirming the contribution of heterogeneous perspective design to classification robustness. The framework currently relies on text input only and provides a basis for extension to multimodal agricultural information.

Lignin-related aromatics are promising renewable feedstocks, but their microbial conversion into useful compounds is fundamentally constrained by substrate toxicity and metabolic flux imbalance. Here, we report an autonomous microbial system in Corynebacterium glutamicum, which integrates evolutionary engineering with multiplexed dynamic control to overcome these constraints. This system features a robust, evolved chassis with reinforced cellular defenses and upgraded aromatic efflux machinery. We engineer a layered, multi-input regulatory network that enables real-time, substrate-responsive and precise control of metabolic pathways. This system achieves 99.89% conversion of diverse lignin-related aromatics to protocatechuic acid and subsequently achieves a titer of 53.40&#x2009;g&#x2009;L&#x207b;&#xb9;. This further supports high-titer production of value-added downstream chemicals, including cis,cis-muconic acid and &#x3b2;-ketoadipic acid with titers of 51.90&#x2009;g&#x2009;L&#x207b;&#xb9; and 38.33&#x2009;g&#x2009;L&#x207b;&#xb9;, respectively. This work establishes a versatile and scalable paradigm for the autonomous bioprocessing of lignin feedstocks into sustainable value-added products.

Large language models (LLMs) show great potential for clinical decision-making, yet most applications remain narrow, task-specific chat tools rather than systems integrated into clinical workflows1,2. However, building physician copilots will require models that operate within the electronic health record (EHR), with governed access to patient data and the ability to initiate permitted EHR actions within defined safety constraints. Yet it remains unproven whether such a system can manage patient cases with physician-level performance. Here we show that MIRA (Medical Intelligence for Reasoning and Action), an autonomous artificial intelligence agent operating in a sandboxed EHR environment, can navigate a large clinical action space to obtain patient histories; order and interpret laboratory, imaging and microbiology tests; generate differential diagnoses; and formulate treatment plans such as prescribing medications, scheduling surgical procedures and planning admissions. In simulations on real patient cases spanning multiple diagnoses, MIRA outperformed physicians in diagnostic accuracy and made guideline-concordant, medication-safe and appropriate admission decisions. Compared with previous LLM applications that addressed isolated subtasks or provided free-text advice, these results suggest that an EHR-integrated artificial intelligence agent can turn clinical intent into structured, actionable EHR operations, possibly making it a more effective decision-support partner for physicians. Further work is needed to establish generalization, safety and governance through prospective, real-world studies.

This study explored the feasibility of combining Raman spectroscopic imaging with data-driven modeling for estimating colchicine release from transdermal patches under in vitro conditions, aiming to reduce the operational complexity and long testing cycle of the conventional paddle-plate method. Ninety representative patch samples were prepared using a Box-Behnken design, with colchicine content, penetration enhancer content, and evaporation time as key variables. Surface Raman imaging data were collected, while reference release profiles were obtained by the paddle-plate method and fitted using the Weibull equation. Based on these data, three models-partial least squares regression, spectra-based convolutional neural network, and image-based convolutional neural network-were developed under curve-fitting-independent and curve-fitting-dependent strategies. Model performance was evaluated using R2, root mean square error, and similarity factors f1 and f2. The curve-fitting-independent strategy showed better predictive performance than the curve-fitting-dependent strategy, and all three models met the commonly used similarity criteria (f1&#x2009;<&#x2009;15 and f2&#x2009;>&#x2009;50). The lower performance of the curve-fitting-dependent strategy was mainly related to scale differences among the release-equation parameters. Green analysis further indicated that the proposed method reduced solvent consumption, waste generation, and energy use compared with conventional testing. Overall, Raman spectroscopic imaging combined with data-driven modeling provides a non-destructive, greener, and relatively rapid approach for in vitro release prediction and quality evaluation of transdermal patches.

This paper presents TriPath3DNet, a novel, efficient, and interpretable 3D CNN architecture designed for real-time, multi-class classification of short, motion-triggered surveillance video clips under challenging real-world conditions-including occlusion, variable lighting, adverse weather, and subtle low-motion anomalies such as loitering or zone intrusion. TriPath3DNet integrates three complementary temporal pathways-short-term motion, long-term context, and Temporal Difference Encoding (TDE)-into a lightweight ResNet3D (R3D)-18 backbone to jointly model transient dynamics and sustained activities. Evaluated on four datasets-including the newly curated Virat1-RC, Virat2-RC, UCF-Crime, and our proprietary In-House Dataset (IHD)-TriPath3DNet achieves state-of-the-art or near state-of-the-art performance, with up to 95.37% accuracy, 99.42% AUC, and an inference latency of 129 to 137&#xa0;ms per 50-frame clip (approx 2.6&#xa0;ms per frame) on an 11&#xa0;GB GPU, using only 33.46&#xa0;M parameters. Notably, it outperforms both CNN- and transformer-based baselines-including MViTv1, MViTv2, and VideoSwin-by significant margins, especially on anomaly-dense benchmarks like UCF-Crime, where most vision transformers struggle. While MViTv2 achieves slightly higher accuracy on IHD (91.87% vs. 89.46%), TriPath3DNet delivers substantially better AUC (98.07% vs. 90.96%), indicating superior calibration for critical anomaly detection. Ablation studies confirm that each temporal branch contributes meaningfully to performance, and Grad-CAM visualizations demonstrate spatially precise and temporally coherent attention maps. By aligning architectural design with edge-cloud deployment constraints and the operational realities of industrial surveillance, our work bridges the gap between academic research and real-world video analytics.