Next Generation Technology (United States)

This paper is part of the ACE2025 Conference Proceedings (View) Full Paper View / Download the full paper in a new tab/window

Modern nutrition science still lacks a comprehensive, machine-readable map linking diet to molecular composition and biological effects. Here we present FoodAtlas, a large-scale knowledge graph that links 1430 foods to 3610 chemicals, 2181 diseases, and 958 flavor descriptors through 96,981 provenance-tracked edges. A transformer-based text-mining pipeline extracted 48,474 quantitative food-chemical associations from 125,723 literature sentences (F1 = 0.67) and integrated them with 23,211 chemical-disease assertions from the Comparative Toxicogenomics Database, 15,222 chemical-bioactivity records from ChEMBL, 3645 flavor annotations from FlavorDB and PubChem, and 6429 taxonomic relationships. Graph embeddings revealed six dietary modules whose signature metabolites delineate distinct, multisystem disease-risk trajectories. Models built on FoodAtlas demonstrate practical utility: a bioactivity predictor achieved strong correlation with antioxidant assays (R² = 0.52; ρ = 0.72), and a substitution engine reduced simulated total disease risk by 11.9%.

ObjectivesThis is a protocol for a Cochrane Review (intervention). The objectives are as follows:To evaluate the benefits and harms of midline catheters versus PICCs for intravenous therapy in adults.

Abstract The performance of network intrusion detection systems is constrained by factors related to data quality and the richness of feature information. Generic data preprocessing methods often struggle to effectively capture local features, which further diminishes their capacity to identify nonlinear and complex intrusion behaviors. Here, an optoelectronic memristor array with multiwavelength control based on oxygen‐rich and oxygen‐deficient amorphous tungsten oxide (WO x (OR)/WO x (OD)) is developed. The device demonstrated synaptic behavior under electronic stimulation. Interestingly, the device shows fast photoresponse and persistent photoconductivity behavior at visible and UV irradiation, respectively, through the modulation of oxygen vacancy ionization and de‐ionization processes in WO x (OD). Leveraging this phenomenon, a localized image enhancement strategy is effectively demonstrated on the device, showcasing substantial improvements in image contrast observed before and after the enhancement process. Furthermore, the localized enhancement approach, applied to the KDDCUP‐99 dataset, enables the convolutional neural network to better capture complex intrusion features, improving detection accuracy from 89.66% to 94.03%. This integrated approach introduces a novel technical pathway for the field of intrusion detection, seamlessly bridging image processing, neuromorphic computing, and cybersecurity applications.

, , , . ,

Background: Male infertility is a common reproductive disorder, affecting about 7% of men in the general population. Despite its prevalence, the cause of infertility is often unknown. This case report presents the results of a comprehensive evaluation of a patient with severe oligoasthenoteratozoospermia and primary infertility. Methods: The patient underwent clinical, andrological, and genetic examinations, including semen analysis, transmission electron microscopy, cytogenetic examination, molecular analysis of the AZF locus and the CFTR gene, whole-exome sequencing, and Sanger sequencing. Results: Semen analysis revealed severe oligoasthenoteratozoospermia. Transmission electron microscopy showed acrosome detachment from the nucleus in 49% of the spermatozoa. A high percentage (54%) of spermatozoa with insufficiently condensed (“immature”) chromatin was also observed. No chromosomal abnormalities, Y chromosome microdeletions, or pathogenic CFTR gene variants were identified. Whole-exome sequencing revealed a novel c.821G>C variant (chrX:127185365G>C; NM_138289.4) in the ACTRT1 gene (Xq25). This variant was hemizygous in the patient and heterozygous in his mother, as determined by Sanger sequencing. According to the ACMG guidelines (PM2, PP3), this missense variant in the ACTRT1 gene was classified as a variant of uncertain clinical significance (VUS). Amino acid conservation and 3D protein modeling predict that the identified variant has a deleterious effect on the protein. Conclusions: This study suggests a potential link between a novel ACTRT1 variant and a specific teratozoospermia phenotype. Further functional studies are needed to confirm this association and determine the role of the gene in X-linked male infertility.

Cryptic species offer opportunities to reveal the mechanisms that constrain phenotypic divergence during speciation. We integrated whole-genome sequencing, morphological, micro- and macro-climatic, and behavioral data to investigate divergence across a well-documented genetic break in the desert-adapted side-blotched lizard, Uta stansburiana, on the Baja California peninsula. Despite deep genomic differentiation, clades show remarkable similarity in morphology, habitat use, and thermal biology. Nearly all genetic differentiation (87%) is explained by isolation by distance and seasonal variation in precipitation, with almost no effect of temperature. Behavioral thermoregulation and changes in activity time accommodate strong macro- and micro-climatic differences, buffering against selection that would otherwise drive morphological and physiological divergence. In contrast, genomic signatures of selection and divergence in genes associated with the nervous system, sensory perception, and biomolecule metabolism indicate adaptation to differences in rainfall seasonality. The results show behavioral flexibility can constrain phenotypic divergence, yielding cryptic species-level genetic divergence despite strong eco-climatic disparities and selection pressures. More broadly, this study shows how rigorous statistical integration of multiple data types can disentangle competing eco-climatic drivers that can decouple phenotype from genotype during speciation.

Large Language Models (LLMs), built on Transformer architectures, exhibit remarkable generalization across a wide range of tasks. However, fine-tuning these models for specific tasks remains resource-intensive due to their extensive parameterization. In this paper, we explore two remarkable phenomena related to the attention mechanism during the fine-tuning of LLMs (where Wq, Wk, and Wv denote the weights of the query, key, and value layers, respectively). The first phenomenon, termed “Unequal Importance of Attention Matrices”, highlights the impact of fine-tuning different weight matrices. It shows that optimizing the Wv matrix yields significantly better performance than optimizing the Wk matrix. Fine-tuning only the Wq and Wv matrices is computationally efficient while delivering results comparable to, or even better than fine-tuning all three matrices (Wq, Wk, and Wv). The second phenomenon, “Attention Matrices with Customized Learning Rate Lead to Better Convergence”, emphasizes the importance of assigning distinct learning rates to these matrices. Specifically, a higher learning rate for the Wv matrix compared to Wq and Wk accelerates convergence and improves performance. Building on these insights, we propose a new strategy that improves fine-tuning efficiency in terms of both storage and time. Experimental results on benchmark datasets validate the effectiveness of this approach, supporting our theoretical findings. Our analysis lays the theoretical groundwork for configuring and improving algorithms in LLMs fine-tuning.

We present a programmable nonvolatile convolutional core based on sub-wavelength phase-change digital meta-surfaces, designed to enable energy-efficient and scalable optical computing. Utilizing sub-wavelength Sb2Se3 cylindrical arrays, the kernel achieves enhanced weight distinguishability, reduced insertion loss, and fine-tuned multi-level reconfigurability, addressing the requirements of optical neural networks (ONNs). Experimental results further demonstrate a weight range from 0.1 to 0.93 with over six levels per unit, supported by a zero static power and digital architecture. We validate the system's 2-bit digital reconfigurability and post-trimming functionality using laser direct-writing techniques. This novel, to the best of our knowledge, integration of phase change materials and digital architectures represents a promising pathway for high-accuracy optical computing applications, such as neural network-based image recognition.

The immense potential of optical vortex arrays (OVAs) has been validated in applications such as particle manipulation, optical imaging, metrology, quantum, and optical communication. Photonic integrated technology has provided a powerful platform for the manipulation of vortex beams, which has the advantages of compact footprint, high controllability, and flexibility. However, since vortex beams are not eigenmodes of rectangular optical waveguides, the on-chip generation of in-plane OVAs has not been reported yet. Here, we propose a controlled method to generate in-plane OVAs by superimposing polarization-multiplexed modes in multimode silicon waveguides. Based on the proposed method, two distinct types of OVAs can be flexibly generated over an ultra-wide optical bandwidth. One exhibits opposite topological charge directions between adjacent vortices (vortex-antivortex), which are achieved by superimposing the fundamental transverse magnetic (TM0) mode with the i-order transverse electric (TEi) mode. The other shows consistent topological charge directions between neighboring vortices, which is realized by superimposing the i-order transverse magnetic (TMi) mode with the TEi mode. By controlling the phase of modes and waveguide width, the topological charge, transmission state, and quantity of vortex beams can be flexibly adjusted.