Search

Artificial intelligence for graph‐structured data has achieved remarkable success in applications such as recommendation systems, social networks, drug discovery, and circuit annotation. Graph convolutional networks (GCNs) are an effective way to learn representations of various graphs. The increasing size and complexity of graphs call for in‐memory computing (IMC) accelerators for GCN to alleviate massive data transmission between off‐chip memory and processing units. However, GCN implementation with IMC is challenging because of the large memory consumption, irregular memory access, and device nonidealities. Herein, a fully binarized GCN (BGCN) accelerator based on computational resistive random‐access memory (RRAM) through software–hardware codesign is presented. The essential operations including aggregation and combination in GCN are implemented on the RRAM crossbar arrays with cooperation between multiply‐and‐accumulation and content‐addressable memory operations. By leveraging the model quantization and IMC on the RRAM, the BGCN accelerator demonstrates less RRAM usage, high robustness to the device variations, high energy efficiency, and comparable classification accuracy compared to the current state‐of‐the‐art GCN accelerators on both graph classification task using the MUTAG and PTC datasets and node classification task using the Cora and CiteSeer datasets. These results provide a promising approach for edge intelligent systems to efficiently process graph‐structured data.

Abstract Combinatorial optimization (CO) has a broad range of applications in various fields, including operations research, computer science, and artificial intelligence. However, many of these problems are classified as nondeterministic polynomial‐time (NP)‐complete or NP‐hard problems, which are known for their computational complexity and cannot be solved in polynomial time on traditional digital computers. To address this challenge, continuous‐time Ising machine solvers have been developed, utilizing different physical principles to map CO problems to ground state finding. However, most Ising machine prototypes operate at speeds comparable to digital hardware and rely on binarizing node states, resulting in increased system complexity and further limiting operating speed. To tackle these issues, a novel device‐algorithm co‐design method is proposed for fast sub‐optimal solution finding with low hardware complexity. On the device side, a piezoelectric lithium niobate (LiNbO3) microelectromechanical system (MEMS) oscillator network‐based Ising machine without second‐harmonic injection locking (SHIL) is devised to solve Max‐cut and graph coloring problems. The LiNbO3 oscillator operates at speeds greater than 9 GHz, making it one of the fastest oscillatory Ising machines. System‐wise, an innovative grouping method is used that achieves a performance guarantee of 0.878 for Max‐cut and 0.658 for graph coloring problems, which is comparable to Ising machines that utilize binarization.

Background The rapid evolution of artificial intelligence (AI) in conjunction with recent updates in dual antiplatelet therapy (DAPT) management guidelines emphasizes the necessity for innovative models to predict ischemic or bleeding events after drug‐eluting stent implantation. Leveraging AI for dynamic prediction has the potential to revolutionize risk stratification and provide personalized decision support for DAPT management. Methods and Results We developed and validated a new AI‐based pipeline using retrospective data of drug‐eluting stent‐treated patients, sourced from the Cerner Health Facts data set (n=98 236) and Optum's de‐identified Clinformatics Data Mart Database (n=9978). The 36 months following drug‐eluting stent implantation were designated as our primary forecasting interval, further segmented into 6 sequential prediction windows. We evaluated 5 distinct AI algorithms for their precision in predicting ischemic and bleeding risks. Model discriminative accuracy was assessed using the area under the receiver operating characteristic curve, among other metrics. The weighted light gradient boosting machine stood out as the preeminent model, thus earning its place as our AI‐DAPT model. The AI‐DAPT demonstrated peak accuracy in the 30 to 36 months window, charting an area under the receiver operating characteristic curve of 90% [95% CI, 88%–92%] for ischemia and 84% [95% CI, 82%–87%] for bleeding predictions. Conclusions Our AI‐DAPT excels in formulating iterative, refined dynamic predictions by assimilating ongoing updates from patients' clinical profiles, holding value as a novel smart clinical tool to facilitate optimal DAPT duration management with high accuracy and adaptability.

Abstract Reflective tomography Lidar is long‐range, high‐resolution imaging Lidar. Because the angular resolution is independent of detection range, it enjoys promising application prospects in imaging of small space targets, estimation of barycentre range of space debris, and many other fields. In practice, images generated by reflective tomography Lidar generally contain a large number of artefacts and noise that need to be removed to obtain the target profile. To improve the quality of the target profile, an algorithm is proposed for the extraction and segmentation of the target region in reflective tomography Lidar images. According to the experimental results, the algorithm can achieve better segmentation results than the traditional threshold segmentation algorithms. In particular, the algorithm can maintain good segmentation results for those images with noticeable ring artefacts, strip artefacts, and noise while avoiding under‐segmentation or over‐segmentation. It also guarantees the integrity of the target segmentation, preserves the outer contour and detailed structure information of the target as much as possible, and improves the accuracy of the target segmentation. Compared with conventional threshold segmentation algorithms, the algorithm improves the quality of image segmentation, and can improve the quality factor by more than 3%.

Abstract In order to overcome the shortcomings of traditional methods for corrosive sulfur detection in transformer oil, Raman spectroscopy based detection is proposed in this paper. The widely concerned corrosive sulfur, Dibenzyl Disulfide (DBDS), was chosen as the characteristic molecule to be detected. A series of oil samples with different DBDS concentrations was prepared. And these samples were extracted by 1‐methyl‐2‐pyrrolidinone for DBDS enrichment to improve the sensitivity of detection. Then Raman spectra of samples were obtained, and a linear model was established by analysing the relationship between the characteristic peak and the DBDS concentration. The limit of detection reached 7.98 mg/kg. For determining the DBDS concentrations causing sulfur corrosion, the sulfur weight content on the copper conductor surface was measured by Scanning Electron Microscope‐Energy Dispersive Spectrometer after a corrosion test. The results show that the corrosion limitation highly depends on the type of transformer oil, and the Raman spectroscopy detection can meet the limit of detection requirement in practical condition. Finally, an on‐site oil sample and five Lab‐made samples were detected via our new method and the current Gas Chromatography‐Mass Spectrometry based method. It is found that there is no significant divergence between the measurement results. And good applicability was also demonstrated in on‐site sample test.

Real‐world time to treatment discontinuation (rwTTD) is an important endpoint measurement of drug efficacy evaluated using real‐world observational data. rwTTD, represented as a set of metrics calculated from a population‐wise curve, cannot be predicted by existing machine learning approaches. Herein, a methodology that enables predicting rwTTD is developed. First, the robust performance of the model in predicting rwTTD across populations of similar or distinct properties with simulated data using a variety of commonly used base learners in machine learning is demonstrated. Then, the robust performance of the approach both within‐cohort and cross‐disease using real‐world observational data of pembrolizumab for advanced lung cancer and head neck cancer is demonstrated. This study establishes a generic pipeline for real‐world time on treatment prediction, which can be extended to any base machine learners and drugs. Currently, there is no existing machine learning approach established for predicting population‐wise rwTTD, despite that it is an essential metric to report real‐world drug efficacy. Therefore, we believe our study opens a new investigation area of rwTTD prediction, and provides an innovative approach to probe this problem and other problems involving population‐wise predictions. An interactive preprint version of the article can be found at: https://doi.org/10.22541/au.166065465.59798123/v1.

Abstract A critical step to evaluate the potential in vivo antiviral activity of a drug is to connect the in vivo exposure to its in vitro antiviral activity. The Anti–SARS‐CoV‐2 Repurposing Drug Database is a database that includes both in vitro anti–SARS‐CoV‐2 activity and in vivo pharmacokinetic data to facilitate the extrapolation from in vitro antiviral activity to potential in vivo antiviral activity for a large set of drugs/compounds. In addition to serving as a data source for in vitro anti–SARS‐CoV‐2 activity and in vivo pharmacokinetic information, the database is also a calculation tool that can be used to compare the in vitro antiviral activity with in vivo drug exposure to identify potential anti–SARS‐CoV‐2 drugs. Continuous development and expansion are feasible with the public availability of this database.

The use of computational models in drug development has grown during the past decade. These model‐informed drug development (MIDD) approaches can inform a variety of drug development and regulatory decisions. When used for regulatory decision making, it is important to establish that the model is credible for its intended use. Currently, there is no consensus on how to establish and assess model credibility, including the selection of appropriate verification and validation activities. In this article, we apply a risk‐informed credibility assessment framework to physiologically‐based pharmacokinetic modeling and simulation and hypothesize this evidentiary framework may also be useful for evaluating other MIDD approaches. We seek to stimulate a scientific discussion around this framework as a potential starting point for uniform assessment of model credibility across MIDD. Ultimately, an overarching framework may help to standardize regulatory evaluation across therapeutic products (i.e., drugs and medical devices).

Background Night eating has been associated with an elevated risk of obesity, dyslipidemia, and cardiovascular disease. However, there is no longitudinal study on whether habitual night eating, regardless of diet quality and energy intake, is associated with arterial stiffness, a major etiological factor in the development of cardiovascular disease. Methods and Results The study included 7771 adult participants without cardiovascular disease, cancer, or diabetes mellitus prior to dietary assessment by a validated food frequency questionnaire in 2014 through 2015. Participants were categorized into 3 groups based on self‐reported night‐eating habits: never or rarely, some days (1–5 times per week), or most days (6+ times per week). Arterial stiffness was assessed by brachial‐ankle pulse wave velocity at baseline and repeatedly during follow‐ups. Mean differences and 95% CIs in the yearly change rate of brachial‐ankle pulse wave velocity across the 3 groups were calculated, adjusting for age, sex, socioeconomic status, total energy intake, diet quality, sleep quality, and other cardiovascular disease risk factors. At baseline, 6625 (85.2%), 610 (7.8%), and 536 (6.9%) participants reported night eating as never or rarely, some days, or most days, respectively. During a mean 3.19 years, we observed a positive association between night‐eating frequency and progression of arterial stiffness (P trend=0.01). The adjusted difference in brachial‐ankle pulse wave velocity change rate between the group that ate at night most days and the group that never or rarely ate at night was 14.1 (95% CI, 0.6–27.5) cm/s per year. This association was only significant in women, but not in men (P interaction=0.03). Conclusions In an adult population free of major chronic diseases, habitual night eating was positively associated with the progression of arterial stiffness, a hallmark of arteriosclerosis and biological aging. Registration URL: https://www.chictr.org.cn; Unique identifier: ChiCTR‐TNRC‐11001489.

Background Elevated long‐term systolic blood pressure and resting heart rate (RHR) variability are suggested to amplify the risk of all‐cause mortality (ACM). However, the link between increased RHR and pulse pressure for ACM remained unclear. Methods and Results This study analyzed 46 751 individuals from Kailuan Cohort Study for the end outcome of ACM. A Cox regression model was used to estimate hazard ratios for death events. Kaplan‐Meier analysis was performed to study the differences in survival as stratified by the SD, coefficient of variation, and average real variability of RHR and pulse pressure quartiles. A total of 1667 deaths (65 years of age had a higher risk for ACM across quartiles of RHR‐SD. The hazard ratio (95% CI) for the subjects in quartiles 2, 3, and 4 were 1.81 (1.10–2.97), 2.31 (1.37–1.3.90), and 2.64 (1.63–4.29), respectively. Conclusions An elevated long‐term RHR variability combined with an increased pulse pressure variability or vice versa amplifies the risk of ACM.