The high porosity and tunable chemical functionality of metal-organic frameworks(MOFs)make it a promising catalyst design platform.High-throughput screening of catalytic performance is feasible since the large MOF str...The high porosity and tunable chemical functionality of metal-organic frameworks(MOFs)make it a promising catalyst design platform.High-throughput screening of catalytic performance is feasible since the large MOF structure database is available.In this study,we report a machine learning model for high-throughput screening of MOF catalysts for the CO_(2) cycloaddition reaction.The descriptors for model training were judiciously chosen according to the reaction mechanism,which leads to high accuracy up to 97%for the 75%quantile of the training set as the classification criterion.The feature contribution was further evaluated with SHAP and PDP analysis to provide a certain physical understanding.12,415 hypothetical MOF structures and 100 reported MOFs were evaluated under 100℃ and 1 bar within one day using the model,and 239 potentially efficient catalysts were discovered.Among them,MOF-76(Y)achieved the top performance experimentally among reported MOFs,in good agreement with the prediction.展开更多
Federated learning(FL)is a distributed machine learning paradigm for edge cloud computing.FL can facilitate data-driven decision-making in tactical scenarios,effectively addressing both data volume and infrastructure ...Federated learning(FL)is a distributed machine learning paradigm for edge cloud computing.FL can facilitate data-driven decision-making in tactical scenarios,effectively addressing both data volume and infrastructure challenges in edge environments.However,the diversity of clients in edge cloud computing presents significant challenges for FL.Personalized federated learning(pFL)received considerable attention in recent years.One example of pFL involves exploiting the global and local information in the local model.Current pFL algorithms experience limitations such as slow convergence speed,catastrophic forgetting,and poor performance in complex tasks,which still have significant shortcomings compared to the centralized learning.To achieve high pFL performance,we propose FedCLCC:Federated Contrastive Learning and Conditional Computing.The core of FedCLCC is the use of contrastive learning and conditional computing.Contrastive learning determines the feature representation similarity to adjust the local model.Conditional computing separates the global and local information and feeds it to their corresponding heads for global and local handling.Our comprehensive experiments demonstrate that FedCLCC outperforms other state-of-the-art FL algorithms.展开更多
It is of great significance to accurately and rapidly identify shale lithofacies in relation to the evaluation and prediction of sweet spots for shale oil and gas reservoirs.To address the problem of low resolution in...It is of great significance to accurately and rapidly identify shale lithofacies in relation to the evaluation and prediction of sweet spots for shale oil and gas reservoirs.To address the problem of low resolution in logging curves,this study establishes a grayscale-phase model based on high-resolution grayscale curves using clustering analysis algorithms for shale lithofacies identification,working with the Shahejie For-mation,Bohai Bay Basin,China.The grayscale phase is defined as the sum of absolute grayscale and relative amplitude as well as their features.The absolute grayscale is the absolute magnitude of the gray values and is utilized for evaluating the material composition(mineral composition+total organic carbon)of shale,while the relative amplitude is the difference between adjacent gray values and is used to identify the shale structure type.The research results show that the grayscale phase model can identify shale lithofacies well,and the accuracy and applicability of this model were verified by the fitting relationship between absolute grayscale and shale mineral composition,as well as corresponding re-lationships between relative amplitudes and laminae development in shales.Four lithofacies are iden-tified in the target layer of the study area:massive mixed shale,laminated mixed shale,massive calcareous shale and laminated calcareous shale.This method can not only effectively characterize the material composition of shale,but also numerically characterize the development degree of shale laminae,and solve the problem that difficult to identify millimeter-scale laminae based on logging curves,which can provide technical support for shale lithofacies identification,sweet spot evaluation and prediction of complex continental lacustrine basins.展开更多
Recent studies employing deep learning to solve the traveling salesman problem(TSP)have mainly focused on learning construction heuristics.Such methods can improve TSP solutions,but still depend on additional programs...Recent studies employing deep learning to solve the traveling salesman problem(TSP)have mainly focused on learning construction heuristics.Such methods can improve TSP solutions,but still depend on additional programs.However,methods that focus on learning improvement heuristics to iteratively refine solutions remain insufficient.Traditional improvement heuristics are guided by a manually designed search strategy and may only achieve limited improvements.This paper proposes a novel framework for learning improvement heuristics,which automatically discovers better improvement policies for heuristics to iteratively solve the TSP.Our framework first designs a new architecture based on a transformer model to make the policy network parameterized,which introduces an action-dropout layer to prevent action selection from overfitting.It then proposes a deep reinforcement learning approach integrating a simulated annealing mechanism(named RL-SA)to learn the pairwise selected policy,aiming to improve the 2-opt algorithm's performance.The RL-SA leverages the whale optimization algorithm to generate initial solutions for better sampling efficiency and uses the Gaussian perturbation strategy to tackle the sparse reward problem of reinforcement learning.The experiment results show that the proposed approach is significantly superior to the state-of-the-art learning-based methods,and further reduces the gap between learning-based methods and highly optimized solvers in the benchmark datasets.Moreover,our pre-trained model M can be applied to guide the SA algorithm(named M-SA(ours)),which performs better than existing deep models in small-,medium-,and large-scale TSPLIB datasets.Additionally,the M-SA(ours)achieves excellent generalization performance in a real-world dataset on global liner shipping routes,with the optimization percentages in distance reduction ranging from3.52%to 17.99%.展开更多
AIM:To establish pupil diameter measurement algorithms based on infrared images that can be used in real-world clinical settings.METHODS:A total of 188 patients from outpatient clinic at He Eye Specialist Shenyang Hos...AIM:To establish pupil diameter measurement algorithms based on infrared images that can be used in real-world clinical settings.METHODS:A total of 188 patients from outpatient clinic at He Eye Specialist Shenyang Hospital from Spetember to December 2022 were included,and 13470 infrared pupil images were collected for the study.All infrared images for pupil segmentation were labeled using the Labelme software.The computation of pupil diameter is divided into four steps:image pre-processing,pupil identification and localization,pupil segmentation,and diameter calculation.Two major models are used in the computation process:the modified YoloV3 and Deeplabv 3+models,which must be trained beforehand.RESULTS:The test dataset included 1348 infrared pupil images.On the test dataset,the modified YoloV3 model had a detection rate of 99.98% and an average precision(AP)of 0.80 for pupils.The DeeplabV3+model achieved a background intersection over union(IOU)of 99.23%,a pupil IOU of 93.81%,and a mean IOU of 96.52%.The pupil diameters in the test dataset ranged from 20 to 56 pixels,with a mean of 36.06±6.85 pixels.The absolute error in pupil diameters between predicted and actual values ranged from 0 to 7 pixels,with a mean absolute error(MAE)of 1.06±0.96 pixels.CONCLUSION:This study successfully demonstrates a robust infrared image-based pupil diameter measurement algorithm,proven to be highly accurate and reliable for clinical application.展开更多
It is known that the commonly used NaSch cellular automaton （CA） model and its modifications can help explain the internal causes of the macro phenomena of traffic flow. However, the randomization probability of veh...It is known that the commonly used NaSch cellular automaton （CA） model and its modifications can help explain the internal causes of the macro phenomena of traffic flow. However, the randomization probability of vehicle velocity used in these models is assumed to be an exogenous constant or a conditional constant, which cannot reflect the learning and forgetting behaviour of drivers with historical experiences. This paper further modifies the NaSch model by enabling the randomization probability to be adjusted on the bases of drivers＇ memory. The Markov properties of this modified model are discussed. Analytical and simulation results show that the traffic fundamental diagrams can be indeed improved when considering drivers＇ intelligent behaviour. Some new features of traffic are revealed by differently combining the model parameters representing learning and forgetting behaviour.展开更多
With the rapid development of the Internet,network security and data privacy are increasingly valued.Although classical Network Intrusion Detection System(NIDS)based on Deep Learning(DL)models can provide good detecti...With the rapid development of the Internet,network security and data privacy are increasingly valued.Although classical Network Intrusion Detection System(NIDS)based on Deep Learning(DL)models can provide good detection accuracy,but collecting samples for centralized training brings the huge risk of data privacy leakage.Furthermore,the training of supervised deep learning models requires a large number of labeled samples,which is usually cumbersome.The“black-box”problem also makes the DL models of NIDS untrustworthy.In this paper,we propose a trusted Federated Learning(FL)Traffic IDS method called FL-TIDS to address the above-mentioned problems.In FL-TIDS,we design an unsupervised intrusion detection model based on autoencoders that alleviates the reliance on marked samples.At the same time,we use FL for model training to protect data privacy.In addition,we design an improved SHAP interpretable method based on chi-square test to perform interpretable analysis of the trained model.We conducted several experiments to evaluate the proposed FL-TIDS.We first determine experimentally the structure and the number of neurons of the unsupervised AE model.Secondly,we evaluated the proposed method using the UNSW-NB15 and CICIDS2017 datasets.The exper-imental results show that the unsupervised AE model has better performance than the other 7 intrusion detection models in terms of precision,recall and f1-score.Then,federated learning is used to train the intrusion detection model.The experimental results indicate that the model is more accurate than the local learning model.Finally,we use an improved SHAP explainability method based on Chi-square test to analyze the explainability.The analysis results show that the identification characteristics of the model are consistent with the attack characteristics,and the model is reliable.展开更多
By using the numerical renormalization group(NRG)method,we construct a large dataset with about one million spectral functions of the Anderson quantum impurity model.The dataset contains the density of states(DOS)of t...By using the numerical renormalization group(NRG)method,we construct a large dataset with about one million spectral functions of the Anderson quantum impurity model.The dataset contains the density of states(DOS)of the host material,the strength of Coulomb interaction between on-site electrons(U),and the hybridization between the host material and the impurity site(Γ).The continued DOS and spectral functions are stored with Chebyshev coefficients and wavelet functions,respectively.From this dataset,we build seven different machine learning networks to predict the spectral function from the input data,DOS,U,andΓ.Three different evaluation indexes,mean absolute error(MAE),relative error(RE)and root mean square error(RMSE),are used to analyze the prediction abilities of different network models.Detailed analysis shows that,for the two kinds of widely used recurrent neural networks(RNNs),gate recurrent unit(GRU)has better performance than the long short term memory(LSTM)network.A combination of bidirectional GRU(BiGRU)and GRU has the best performance among GRU,BiGRU,LSTM,and BiLSTM.The MAE peak of BiGRU+GRU reaches 0.00037.We have also tested a one-dimensional convolutional neural network(1DCNN)with 20 hidden layers and a residual neural network(ResNet),we find that the 1DCNN has almost the same performance of the BiGRU+GRU network for the original dataset,while the robustness testing seems to be a little weak than BiGRU+GRU when we test all these models on two other independent datasets.The ResNet has the worst performance among all the seven network models.The datasets presented in this paper,including the large data set of the spectral function of Anderson quantum impurity model,are openly available at https://doi.org/10.57760/sciencedb.j00113.00192.展开更多
Forest habitats are critical for biodiversity,ecosystem services,human livelihoods,and well-being.Capacity to conduct theoretical and applied forest ecology research addressing direct(e.g.,deforestation)and indirect(e...Forest habitats are critical for biodiversity,ecosystem services,human livelihoods,and well-being.Capacity to conduct theoretical and applied forest ecology research addressing direct(e.g.,deforestation)and indirect(e.g.,climate change)anthropogenic pressures has benefited considerably from new field-and statistical-techniques.We used machine learning and bibliometric structural topic modelling to identify 20 latent topics comprising four principal fields from a corpus of 16,952 forest ecology/forestry articles published in eight ecology and five forestry journals between 2010 and 2022.Articles published per year increased from 820 in 2010 to 2,354 in 2021,shifting toward more applied topics.Publications from China and some countries in North America and Europe dominated,with relatively fewer articles from some countries in West and Central Africa and West Asia,despite globally important forest resources.Most study sites were in some countries in North America,Central Asia,and South America,and Australia.Articles utilizing R statistical software predominated,increasing from 29.5%in 2010 to 71.4%in 2022.The most frequently used packages included lme4,vegan,nlme,MuMIn,ggplot2,car,MASS,mgcv,multcomp and raster.R was more often used in forest ecology than applied forestry articles.R software offers advantages in script and workflow-sharing compared to other statistical packages.Our findings demonstrate that the disciplines of forest ecology/forestry are expanding both in number and scope,aided by more sophisticated statistical tools,to tackle the challenges of redressing forest habitat loss and the socio-economic impacts of deforestation.展开更多
Machine learning-based surrogate models have significant advantages in terms of computing efficiency. In this paper, we present a pilot study on fast calibration using machine learning techniques. Technology computer-...Machine learning-based surrogate models have significant advantages in terms of computing efficiency. In this paper, we present a pilot study on fast calibration using machine learning techniques. Technology computer-aided design(TCAD) is a powerful simulation tool for electronic devices. This simulation tool has been widely used in the research of radiation effects.However, calibration of TCAD models is time-consuming. In this study, we introduce a fast calibration approach for TCAD model calibration of metal–oxide–semiconductor field-effect transistors(MOSFETs). This approach utilized a machine learning-based surrogate model that was several orders of magnitude faster than the original TCAD simulation. The desired calibration results were obtained within several seconds. In this study, a fundamental model containing 26 parameters is introduced to represent the typical structure of a MOSFET. Classifications were developed to improve the efficiency of the training sample generation. Feature selection techniques were employed to identify important parameters. A surrogate model consisting of a classifier and a regressor was built. A calibration procedure based on the surrogate model was proposed and tested with three calibration goals. Our work demonstrates the feasibility of machine learning-based fast model calibrations for MOSFET. In addition, this study shows that these machine learning techniques learn patterns and correlations from data instead of employing domain expertise. This indicates that machine learning could be an alternative research approach to complement classical physics-based research.展开更多
The safety assessment of high-level radioactive waste repositories requires a high predictive accuracy for radionuclide diffusion and a comprehensive understanding of the diffusion mechanism.In this study,a through-di...The safety assessment of high-level radioactive waste repositories requires a high predictive accuracy for radionuclide diffusion and a comprehensive understanding of the diffusion mechanism.In this study,a through-diffusion method and six machine-learning methods were employed to investigate the diffusion of ReO_(4)^(−),HCrO_(4)^(−),and I−in saturated compacted bentonite under different salinities and compacted dry densities.The machine-learning models were trained using two datasets.One dataset contained six input features and 293 instances obtained from the diffusion database system of the Japan Atomic Energy Agency(JAEA-DDB)and 15 publications.The other dataset,comprising 15,000 pseudo-instances,was produced using a multi-porosity model and contained eight input features.The results indicate that the former dataset yielded a higher predictive accuracy than the latter.Light gradient-boosting exhibited a higher prediction accuracy(R2=0.92)and lower error(MSE=0.01)than the other machine-learning algorithms.In addition,Shapley Additive Explanations,Feature Importance,and Partial Dependence Plot analysis results indicate that the rock capacity factor and compacted dry density had the two most significant effects on predicting the effective diffusion coefficient,thereby offering valuable insights.展开更多
Building information modeling(BIM)object classification takes a lot of time and energy.Misclassification or omission of any object may lead to the emergence of abnormal results,which have a great impact on the project...Building information modeling(BIM)object classification takes a lot of time and energy.Misclassification or omission of any object may lead to the emergence of abnormal results,which have a great impact on the project workflow and results.Roundly understanding BIM object classification,by improving Swin Transformer classifier algorithm parameters,using the model primitives extracted from IFC format BIM model file,deep learning of 7 types of BIM object categories is taken.Through the performance and evaluation indicators obtained in training,the results improve the classification accuracy.展开更多
With the projected global surge in hydrogen demand, driven by increasing applications and the imperative for low-emission hydrogen, the integration of machine learning(ML) across the hydrogen energy value chain is a c...With the projected global surge in hydrogen demand, driven by increasing applications and the imperative for low-emission hydrogen, the integration of machine learning(ML) across the hydrogen energy value chain is a compelling avenue. This review uniquely focuses on harnessing the synergy between ML and computational modeling(CM) or optimization tools, as well as integrating multiple ML techniques with CM, for the synthesis of diverse hydrogen evolution reaction(HER) catalysts and various hydrogen production processes(HPPs). Furthermore, this review addresses a notable gap in the literature by offering insights, analyzing challenges, and identifying research prospects and opportunities for sustainable hydrogen production. While the literature reflects a promising landscape for ML applications in hydrogen energy domains, transitioning AI-based algorithms from controlled environments to real-world applications poses significant challenges. Hence, this comprehensive review delves into the technical,practical, and ethical considerations associated with the application of ML in HER catalyst development and HPP optimization. Overall, this review provides guidance for unlocking the transformative potential of ML in enhancing prediction efficiency and sustainability in the hydrogen production sector.展开更多
The multi-source passive localization problem is a problem of great interest in signal pro-cessing with many applications.In this paper,a sparse representation model based on covariance matrix is constructed for the l...The multi-source passive localization problem is a problem of great interest in signal pro-cessing with many applications.In this paper,a sparse representation model based on covariance matrix is constructed for the long-range localization scenario,and a sparse Bayesian learning algo-rithm based on Laplace prior of signal covariance is developed for the base mismatch problem caused by target deviation from the initial point grid.An adaptive grid sparse Bayesian learning targets localization(AGSBL)algorithm is proposed.The AGSBL algorithm implements a covari-ance-based sparse signal reconstruction and grid adaptive localization dictionary learning.Simula-tion results show that the AGSBL algorithm outperforms the traditional compressed-aware localiza-tion algorithm for different signal-to-noise ratios and different number of targets in long-range scenes.展开更多
Photonic inverse design concerns the problem of finding photonic structures with target optical properties.However,traditional methods based on optimization algorithms are time-consuming and computationally expensive....Photonic inverse design concerns the problem of finding photonic structures with target optical properties.However,traditional methods based on optimization algorithms are time-consuming and computationally expensive.Recently,deep learning-based approaches have been developed to tackle the problem of inverse design efficiently.Although most of these neural network models have demonstrated high accuracy in different inverse design problems,no previous study has examined the potential effects under given constraints in nanomanufacturing.Additionally,the relative strength of different deep learning-based inverse design approaches has not been fully investigated.Here,we benchmark three commonly used deep learning models in inverse design:Tandem networks,Variational Auto-Encoders,and Generative Adversarial Networks.We provide detailed comparisons in terms of their accuracy,diversity,and robustness.We find that tandem networks and Variational Auto-Encoders give the best accuracy,while Generative Adversarial Networks lead to the most diverse predictions.Our findings could serve as a guideline for researchers to select the model that can best suit their design criteria and fabrication considerations.In addition,our code and data are publicly available,which could be used for future inverse design model development and benchmarking.展开更多
The contribution of this work is twofold： （1） a multimodality prediction method of chaotic time series with the Gaussian process mixture （GPM） model is proposed, which employs a divide and conquer strategy. It au...The contribution of this work is twofold： （1） a multimodality prediction method of chaotic time series with the Gaussian process mixture （GPM） model is proposed, which employs a divide and conquer strategy. It automatically divides the chaotic time series into multiple modalities with different extrinsic patterns and intrinsic characteristics, and thus can more precisely fit the chaotic time series. （2） An effective sparse hard-cut expec- tation maximization （SHC-EM） learning algorithm for the GPM model is proposed to improve the prediction performance. SHO-EM replaces a large learning sample set with fewer pseudo inputs, accelerating model learning based on these pseudo inputs. Experiments on Lorenz and Chua time series demonstrate that the proposed method yields not only accurate multimodality prediction, but also the prediction confidence interval SHC-EM outperforms the traditional variational 1earning in terms of both prediction accuracy and speed. In addition, SHC-EM is more robust and insusceptible to noise than variational learning.展开更多
Traditional Numerical Reservoir Simulation has been contributing to the oil and gas industry for decades.The current state of this technology is the result of decades of research and development by a large number of e...Traditional Numerical Reservoir Simulation has been contributing to the oil and gas industry for decades.The current state of this technology is the result of decades of research and development by a large number of engineers and scientists.Starting in the late 1960s and early 1970s,advances in computer hardware along with development and adaptation of clever algorithms resulted in a paradigm shift in reservoir studies moving them from simplified analogs and analytical solution methods to more mathematically robust computational and numerical solution models.展开更多
Action model learning has become a hot topic in knowledge engineering for automated planning.A key problem for learning action models is to analyze state changes before and after action executions from observed"p...Action model learning has become a hot topic in knowledge engineering for automated planning.A key problem for learning action models is to analyze state changes before and after action executions from observed"plan traces".To support such an analysis,a new approach is proposed to partition propositions of plan traces into states.First,vector representations of propositions and actions are obtained by training a neural network called Skip-Gram borrowed from the area of natural language processing(NLP).Then,a type of semantic distance among propositions and actions is defined based on their similarity measures in the vector space.Finally,k-means and k-nearest neighbor(kNN)algorithms are exploited to map propositions to states.This approach is called state partition by word vector(SPWV),which is implemented on top of a recent action model learning framework by Rao et al.Experimental results on the benchmark domains show that SPWV leads to a lower error rate of the learnt action model,compared to the probability based approach for state partition that was developed by Rao et al.展开更多
Conducting hydrodynamic and physical motion simulation tests using a large-scale self-propelled model under actual wave conditions is an important means for researching environmental adaptability of ships. During the ...Conducting hydrodynamic and physical motion simulation tests using a large-scale self-propelled model under actual wave conditions is an important means for researching environmental adaptability of ships. During the navigation test of the self-propelled model, the complex environment including various port facilities, navigation facilities, and the ships nearby must be considered carefully, because in this dense environment the impact of sea waves and winds on the model is particularly significant. In order to improve the security of the self-propelled model, this paper introduces the Q learning based on reinforcement learning combined with chaotic ideas for the model＇s collision avoidance, in order to improve the reliability of the local path planning. Simulation and sea test results show that this algorithm is a better solution for collision avoidance of the self navigation model under the interference of sea winds and waves with good adaptability.展开更多
It is shown that we can control spatiotemporal chaos in the Frenkel-Kontorova(FK)model by a model-free control method based on reinforcement learning.The method uses Q-learning to find optimal control strategies based...It is shown that we can control spatiotemporal chaos in the Frenkel-Kontorova(FK)model by a model-free control method based on reinforcement learning.The method uses Q-learning to find optimal control strategies based on the reward feedback from the environment that maximizes its performance.The optimal control strategies are recorded in a Q-table and then employed to implement controllers.The advantage of the method is that it does not require an explicit knowledge of the system,target states,and unstable periodic orbits.All that we need is the parameters that we are trying to control and an unknown simulation model that represents the interactive environment.To control the FK model,we employ the perturbation policy on two different kinds of parameters,i.e.,the pendulum lengths and the phase angles.We show that both of the two perturbation techniques,i.e.,changing the lengths and changing their phase angles,can suppress chaos in the system and make it create the periodic patterns.The form of patterns depends on the initial values of the angular displacements and velocities.In particular,we show that the pinning control strategy,which only changes a small number of lengths or phase angles,can be put into effect.展开更多
基金financial support from the National Key Research and Development Program of China(2021YFB 3501501)the National Natural Science Foundation of China(No.22225803,22038001,22108007 and 22278011)+1 种基金Beijing Natural Science Foundation(No.Z230023)Beijing Science and Technology Commission(No.Z211100004321001).
文摘The high porosity and tunable chemical functionality of metal-organic frameworks(MOFs)make it a promising catalyst design platform.High-throughput screening of catalytic performance is feasible since the large MOF structure database is available.In this study,we report a machine learning model for high-throughput screening of MOF catalysts for the CO_(2) cycloaddition reaction.The descriptors for model training were judiciously chosen according to the reaction mechanism,which leads to high accuracy up to 97%for the 75%quantile of the training set as the classification criterion.The feature contribution was further evaluated with SHAP and PDP analysis to provide a certain physical understanding.12,415 hypothetical MOF structures and 100 reported MOFs were evaluated under 100℃ and 1 bar within one day using the model,and 239 potentially efficient catalysts were discovered.Among them,MOF-76(Y)achieved the top performance experimentally among reported MOFs,in good agreement with the prediction.
基金supported by the Natural Science Foundation of Xinjiang Uygur Autonomous Region(Grant No.2022D01B 187)。
文摘Federated learning(FL)is a distributed machine learning paradigm for edge cloud computing.FL can facilitate data-driven decision-making in tactical scenarios,effectively addressing both data volume and infrastructure challenges in edge environments.However,the diversity of clients in edge cloud computing presents significant challenges for FL.Personalized federated learning(pFL)received considerable attention in recent years.One example of pFL involves exploiting the global and local information in the local model.Current pFL algorithms experience limitations such as slow convergence speed,catastrophic forgetting,and poor performance in complex tasks,which still have significant shortcomings compared to the centralized learning.To achieve high pFL performance,we propose FedCLCC:Federated Contrastive Learning and Conditional Computing.The core of FedCLCC is the use of contrastive learning and conditional computing.Contrastive learning determines the feature representation similarity to adjust the local model.Conditional computing separates the global and local information and feeds it to their corresponding heads for global and local handling.Our comprehensive experiments demonstrate that FedCLCC outperforms other state-of-the-art FL algorithms.
基金supported by the National Natural Science Foundation of China(42122017,41821002)the Independent Innovation Research Program of China University of Petroleum(East China)(21CX06001A).
文摘It is of great significance to accurately and rapidly identify shale lithofacies in relation to the evaluation and prediction of sweet spots for shale oil and gas reservoirs.To address the problem of low resolution in logging curves,this study establishes a grayscale-phase model based on high-resolution grayscale curves using clustering analysis algorithms for shale lithofacies identification,working with the Shahejie For-mation,Bohai Bay Basin,China.The grayscale phase is defined as the sum of absolute grayscale and relative amplitude as well as their features.The absolute grayscale is the absolute magnitude of the gray values and is utilized for evaluating the material composition(mineral composition+total organic carbon)of shale,while the relative amplitude is the difference between adjacent gray values and is used to identify the shale structure type.The research results show that the grayscale phase model can identify shale lithofacies well,and the accuracy and applicability of this model were verified by the fitting relationship between absolute grayscale and shale mineral composition,as well as corresponding re-lationships between relative amplitudes and laminae development in shales.Four lithofacies are iden-tified in the target layer of the study area:massive mixed shale,laminated mixed shale,massive calcareous shale and laminated calcareous shale.This method can not only effectively characterize the material composition of shale,but also numerically characterize the development degree of shale laminae,and solve the problem that difficult to identify millimeter-scale laminae based on logging curves,which can provide technical support for shale lithofacies identification,sweet spot evaluation and prediction of complex continental lacustrine basins.
基金Project supported by the National Natural Science Foundation of China(Grant Nos.72101046 and 61672128)。
文摘Recent studies employing deep learning to solve the traveling salesman problem(TSP)have mainly focused on learning construction heuristics.Such methods can improve TSP solutions,but still depend on additional programs.However,methods that focus on learning improvement heuristics to iteratively refine solutions remain insufficient.Traditional improvement heuristics are guided by a manually designed search strategy and may only achieve limited improvements.This paper proposes a novel framework for learning improvement heuristics,which automatically discovers better improvement policies for heuristics to iteratively solve the TSP.Our framework first designs a new architecture based on a transformer model to make the policy network parameterized,which introduces an action-dropout layer to prevent action selection from overfitting.It then proposes a deep reinforcement learning approach integrating a simulated annealing mechanism(named RL-SA)to learn the pairwise selected policy,aiming to improve the 2-opt algorithm's performance.The RL-SA leverages the whale optimization algorithm to generate initial solutions for better sampling efficiency and uses the Gaussian perturbation strategy to tackle the sparse reward problem of reinforcement learning.The experiment results show that the proposed approach is significantly superior to the state-of-the-art learning-based methods,and further reduces the gap between learning-based methods and highly optimized solvers in the benchmark datasets.Moreover,our pre-trained model M can be applied to guide the SA algorithm(named M-SA(ours)),which performs better than existing deep models in small-,medium-,and large-scale TSPLIB datasets.Additionally,the M-SA(ours)achieves excellent generalization performance in a real-world dataset on global liner shipping routes,with the optimization percentages in distance reduction ranging from3.52%to 17.99%.
文摘AIM:To establish pupil diameter measurement algorithms based on infrared images that can be used in real-world clinical settings.METHODS:A total of 188 patients from outpatient clinic at He Eye Specialist Shenyang Hospital from Spetember to December 2022 were included,and 13470 infrared pupil images were collected for the study.All infrared images for pupil segmentation were labeled using the Labelme software.The computation of pupil diameter is divided into four steps:image pre-processing,pupil identification and localization,pupil segmentation,and diameter calculation.Two major models are used in the computation process:the modified YoloV3 and Deeplabv 3+models,which must be trained beforehand.RESULTS:The test dataset included 1348 infrared pupil images.On the test dataset,the modified YoloV3 model had a detection rate of 99.98% and an average precision(AP)of 0.80 for pupils.The DeeplabV3+model achieved a background intersection over union(IOU)of 99.23%,a pupil IOU of 93.81%,and a mean IOU of 96.52%.The pupil diameters in the test dataset ranged from 20 to 56 pixels,with a mean of 36.06±6.85 pixels.The absolute error in pupil diameters between predicted and actual values ranged from 0 to 7 pixels,with a mean absolute error(MAE)of 1.06±0.96 pixels.CONCLUSION:This study successfully demonstrates a robust infrared image-based pupil diameter measurement algorithm,proven to be highly accurate and reliable for clinical application.
基金supported by the National Natural Science Foundation of China (Grant No. 70821061)the National Basic Research Program of China (Grant No. 2006CB705503)
文摘It is known that the commonly used NaSch cellular automaton （CA） model and its modifications can help explain the internal causes of the macro phenomena of traffic flow. However, the randomization probability of vehicle velocity used in these models is assumed to be an exogenous constant or a conditional constant, which cannot reflect the learning and forgetting behaviour of drivers with historical experiences. This paper further modifies the NaSch model by enabling the randomization probability to be adjusted on the bases of drivers＇ memory. The Markov properties of this modified model are discussed. Analytical and simulation results show that the traffic fundamental diagrams can be indeed improved when considering drivers＇ intelligent behaviour. Some new features of traffic are revealed by differently combining the model parameters representing learning and forgetting behaviour.
基金supported by National Natural Science Fundation of China under Grant 61972208National Natural Science Fundation(General Program)of China under Grant 61972211+2 种基金National Key Research and Development Project of China under Grant 2020YFB1804700Future Network Innovation Research and Application Projects under Grant No.2021FNA020062021 Jiangsu Postgraduate Research Innovation Plan under Grant No.KYCX210794.
文摘With the rapid development of the Internet,network security and data privacy are increasingly valued.Although classical Network Intrusion Detection System(NIDS)based on Deep Learning(DL)models can provide good detection accuracy,but collecting samples for centralized training brings the huge risk of data privacy leakage.Furthermore,the training of supervised deep learning models requires a large number of labeled samples,which is usually cumbersome.The“black-box”problem also makes the DL models of NIDS untrustworthy.In this paper,we propose a trusted Federated Learning(FL)Traffic IDS method called FL-TIDS to address the above-mentioned problems.In FL-TIDS,we design an unsupervised intrusion detection model based on autoencoders that alleviates the reliance on marked samples.At the same time,we use FL for model training to protect data privacy.In addition,we design an improved SHAP interpretable method based on chi-square test to perform interpretable analysis of the trained model.We conducted several experiments to evaluate the proposed FL-TIDS.We first determine experimentally the structure and the number of neurons of the unsupervised AE model.Secondly,we evaluated the proposed method using the UNSW-NB15 and CICIDS2017 datasets.The exper-imental results show that the unsupervised AE model has better performance than the other 7 intrusion detection models in terms of precision,recall and f1-score.Then,federated learning is used to train the intrusion detection model.The experimental results indicate that the model is more accurate than the local learning model.Finally,we use an improved SHAP explainability method based on Chi-square test to analyze the explainability.The analysis results show that the identification characteristics of the model are consistent with the attack characteristics,and the model is reliable.
基金Project supported by the National Natural Science Foundation of China(Grant No.12174101)the Fundamental Research Funds for the Central Universities(Grant No.2022MS051)。
文摘By using the numerical renormalization group(NRG)method,we construct a large dataset with about one million spectral functions of the Anderson quantum impurity model.The dataset contains the density of states(DOS)of the host material,the strength of Coulomb interaction between on-site electrons(U),and the hybridization between the host material and the impurity site(Γ).The continued DOS and spectral functions are stored with Chebyshev coefficients and wavelet functions,respectively.From this dataset,we build seven different machine learning networks to predict the spectral function from the input data,DOS,U,andΓ.Three different evaluation indexes,mean absolute error(MAE),relative error(RE)and root mean square error(RMSE),are used to analyze the prediction abilities of different network models.Detailed analysis shows that,for the two kinds of widely used recurrent neural networks(RNNs),gate recurrent unit(GRU)has better performance than the long short term memory(LSTM)network.A combination of bidirectional GRU(BiGRU)and GRU has the best performance among GRU,BiGRU,LSTM,and BiLSTM.The MAE peak of BiGRU+GRU reaches 0.00037.We have also tested a one-dimensional convolutional neural network(1DCNN)with 20 hidden layers and a residual neural network(ResNet),we find that the 1DCNN has almost the same performance of the BiGRU+GRU network for the original dataset,while the robustness testing seems to be a little weak than BiGRU+GRU when we test all these models on two other independent datasets.The ResNet has the worst performance among all the seven network models.The datasets presented in this paper,including the large data set of the spectral function of Anderson quantum impurity model,are openly available at https://doi.org/10.57760/sciencedb.j00113.00192.
基金financially supported by the National Natural Science Foundation of China(31971541).
文摘Forest habitats are critical for biodiversity,ecosystem services,human livelihoods,and well-being.Capacity to conduct theoretical and applied forest ecology research addressing direct(e.g.,deforestation)and indirect(e.g.,climate change)anthropogenic pressures has benefited considerably from new field-and statistical-techniques.We used machine learning and bibliometric structural topic modelling to identify 20 latent topics comprising four principal fields from a corpus of 16,952 forest ecology/forestry articles published in eight ecology and five forestry journals between 2010 and 2022.Articles published per year increased from 820 in 2010 to 2,354 in 2021,shifting toward more applied topics.Publications from China and some countries in North America and Europe dominated,with relatively fewer articles from some countries in West and Central Africa and West Asia,despite globally important forest resources.Most study sites were in some countries in North America,Central Asia,and South America,and Australia.Articles utilizing R statistical software predominated,increasing from 29.5%in 2010 to 71.4%in 2022.The most frequently used packages included lme4,vegan,nlme,MuMIn,ggplot2,car,MASS,mgcv,multcomp and raster.R was more often used in forest ecology than applied forestry articles.R software offers advantages in script and workflow-sharing compared to other statistical packages.Our findings demonstrate that the disciplines of forest ecology/forestry are expanding both in number and scope,aided by more sophisticated statistical tools,to tackle the challenges of redressing forest habitat loss and the socio-economic impacts of deforestation.
基金supported by the National Natural Science Foundation of China (Nos. 11690040 and 11690043)。
文摘Machine learning-based surrogate models have significant advantages in terms of computing efficiency. In this paper, we present a pilot study on fast calibration using machine learning techniques. Technology computer-aided design(TCAD) is a powerful simulation tool for electronic devices. This simulation tool has been widely used in the research of radiation effects.However, calibration of TCAD models is time-consuming. In this study, we introduce a fast calibration approach for TCAD model calibration of metal–oxide–semiconductor field-effect transistors(MOSFETs). This approach utilized a machine learning-based surrogate model that was several orders of magnitude faster than the original TCAD simulation. The desired calibration results were obtained within several seconds. In this study, a fundamental model containing 26 parameters is introduced to represent the typical structure of a MOSFET. Classifications were developed to improve the efficiency of the training sample generation. Feature selection techniques were employed to identify important parameters. A surrogate model consisting of a classifier and a regressor was built. A calibration procedure based on the surrogate model was proposed and tested with three calibration goals. Our work demonstrates the feasibility of machine learning-based fast model calibrations for MOSFET. In addition, this study shows that these machine learning techniques learn patterns and correlations from data instead of employing domain expertise. This indicates that machine learning could be an alternative research approach to complement classical physics-based research.
基金the Key Program of National Natural Science Foundation of China(No.12335008),the Postgraduate Research and Innovation Project of Huzhou University(No.2023KYCX62)the Scientific Research Fund of Zhejiang Provincial Education Department(No.Y202352712)the Huzhou science and technology planning project(No.2021GZ60)。
文摘The safety assessment of high-level radioactive waste repositories requires a high predictive accuracy for radionuclide diffusion and a comprehensive understanding of the diffusion mechanism.In this study,a through-diffusion method and six machine-learning methods were employed to investigate the diffusion of ReO_(4)^(−),HCrO_(4)^(−),and I−in saturated compacted bentonite under different salinities and compacted dry densities.The machine-learning models were trained using two datasets.One dataset contained six input features and 293 instances obtained from the diffusion database system of the Japan Atomic Energy Agency(JAEA-DDB)and 15 publications.The other dataset,comprising 15,000 pseudo-instances,was produced using a multi-porosity model and contained eight input features.The results indicate that the former dataset yielded a higher predictive accuracy than the latter.Light gradient-boosting exhibited a higher prediction accuracy(R2=0.92)and lower error(MSE=0.01)than the other machine-learning algorithms.In addition,Shapley Additive Explanations,Feature Importance,and Partial Dependence Plot analysis results indicate that the rock capacity factor and compacted dry density had the two most significant effects on predicting the effective diffusion coefficient,thereby offering valuable insights.
文摘Building information modeling(BIM)object classification takes a lot of time and energy.Misclassification or omission of any object may lead to the emergence of abnormal results,which have a great impact on the project workflow and results.Roundly understanding BIM object classification,by improving Swin Transformer classifier algorithm parameters,using the model primitives extracted from IFC format BIM model file,deep learning of 7 types of BIM object categories is taken.Through the performance and evaluation indicators obtained in training,the results improve the classification accuracy.
基金express their gratitude to the Higher Institution Centre of Excellence (HICoE) fund under the project code (JPT.S(BPKI)2000/016/018/015JId.4(21)/2022002HICOE)Universiti Tenaga Nasional (UNITEN) for funding the research through the (J510050002–IC–6 BOLDREFRESH2025)Akaun Amanah Industri Bekalan Elektrik (AAIBE) Chair of Renewable Energy grant,and NEC Energy Transition Grant (202203003ETG)。
文摘With the projected global surge in hydrogen demand, driven by increasing applications and the imperative for low-emission hydrogen, the integration of machine learning(ML) across the hydrogen energy value chain is a compelling avenue. This review uniquely focuses on harnessing the synergy between ML and computational modeling(CM) or optimization tools, as well as integrating multiple ML techniques with CM, for the synthesis of diverse hydrogen evolution reaction(HER) catalysts and various hydrogen production processes(HPPs). Furthermore, this review addresses a notable gap in the literature by offering insights, analyzing challenges, and identifying research prospects and opportunities for sustainable hydrogen production. While the literature reflects a promising landscape for ML applications in hydrogen energy domains, transitioning AI-based algorithms from controlled environments to real-world applications poses significant challenges. Hence, this comprehensive review delves into the technical,practical, and ethical considerations associated with the application of ML in HER catalyst development and HPP optimization. Overall, this review provides guidance for unlocking the transformative potential of ML in enhancing prediction efficiency and sustainability in the hydrogen production sector.
文摘The multi-source passive localization problem is a problem of great interest in signal pro-cessing with many applications.In this paper,a sparse representation model based on covariance matrix is constructed for the long-range localization scenario,and a sparse Bayesian learning algo-rithm based on Laplace prior of signal covariance is developed for the base mismatch problem caused by target deviation from the initial point grid.An adaptive grid sparse Bayesian learning targets localization(AGSBL)algorithm is proposed.The AGSBL algorithm implements a covari-ance-based sparse signal reconstruction and grid adaptive localization dictionary learning.Simula-tion results show that the AGSBL algorithm outperforms the traditional compressed-aware localiza-tion algorithm for different signal-to-noise ratios and different number of targets in long-range scenes.
文摘Photonic inverse design concerns the problem of finding photonic structures with target optical properties.However,traditional methods based on optimization algorithms are time-consuming and computationally expensive.Recently,deep learning-based approaches have been developed to tackle the problem of inverse design efficiently.Although most of these neural network models have demonstrated high accuracy in different inverse design problems,no previous study has examined the potential effects under given constraints in nanomanufacturing.Additionally,the relative strength of different deep learning-based inverse design approaches has not been fully investigated.Here,we benchmark three commonly used deep learning models in inverse design:Tandem networks,Variational Auto-Encoders,and Generative Adversarial Networks.We provide detailed comparisons in terms of their accuracy,diversity,and robustness.We find that tandem networks and Variational Auto-Encoders give the best accuracy,while Generative Adversarial Networks lead to the most diverse predictions.Our findings could serve as a guideline for researchers to select the model that can best suit their design criteria and fabrication considerations.In addition,our code and data are publicly available,which could be used for future inverse design model development and benchmarking.
基金Supported by the National Natural Science Foundation of China under Grant No 60972106the China Postdoctoral Science Foundation under Grant No 2014M561053+1 种基金the Humanity and Social Science Foundation of Ministry of Education of China under Grant No 15YJA630108the Hebei Province Natural Science Foundation under Grant No E2016202341
文摘The contribution of this work is twofold： （1） a multimodality prediction method of chaotic time series with the Gaussian process mixture （GPM） model is proposed, which employs a divide and conquer strategy. It automatically divides the chaotic time series into multiple modalities with different extrinsic patterns and intrinsic characteristics, and thus can more precisely fit the chaotic time series. （2） An effective sparse hard-cut expec- tation maximization （SHC-EM） learning algorithm for the GPM model is proposed to improve the prediction performance. SHO-EM replaces a large learning sample set with fewer pseudo inputs, accelerating model learning based on these pseudo inputs. Experiments on Lorenz and Chua time series demonstrate that the proposed method yields not only accurate multimodality prediction, but also the prediction confidence interval SHC-EM outperforms the traditional variational 1earning in terms of both prediction accuracy and speed. In addition, SHC-EM is more robust and insusceptible to noise than variational learning.
文摘Traditional Numerical Reservoir Simulation has been contributing to the oil and gas industry for decades.The current state of this technology is the result of decades of research and development by a large number of engineers and scientists.Starting in the late 1960s and early 1970s,advances in computer hardware along with development and adaptation of clever algorithms resulted in a paradigm shift in reservoir studies moving them from simplified analogs and analytical solution methods to more mathematically robust computational and numerical solution models.
基金Supported by the National Natural Science Foundation of China(61103136,61370156,61503074)Open Research Foundation of Science and Technology on Aerospace Flight Dynamics Laboratory(2014afdl002)
文摘Action model learning has become a hot topic in knowledge engineering for automated planning.A key problem for learning action models is to analyze state changes before and after action executions from observed"plan traces".To support such an analysis,a new approach is proposed to partition propositions of plan traces into states.First,vector representations of propositions and actions are obtained by training a neural network called Skip-Gram borrowed from the area of natural language processing(NLP).Then,a type of semantic distance among propositions and actions is defined based on their similarity measures in the vector space.Finally,k-means and k-nearest neighbor(kNN)algorithms are exploited to map propositions to states.This approach is called state partition by word vector(SPWV),which is implemented on top of a recent action model learning framework by Rao et al.Experimental results on the benchmark domains show that SPWV leads to a lower error rate of the learnt action model,compared to the probability based approach for state partition that was developed by Rao et al.
基金Foundation item： Supported by the National Natural Science Foundation of China under Grant No.61100005.
文摘Conducting hydrodynamic and physical motion simulation tests using a large-scale self-propelled model under actual wave conditions is an important means for researching environmental adaptability of ships. During the navigation test of the self-propelled model, the complex environment including various port facilities, navigation facilities, and the ships nearby must be considered carefully, because in this dense environment the impact of sea waves and winds on the model is particularly significant. In order to improve the security of the self-propelled model, this paper introduces the Q learning based on reinforcement learning combined with chaotic ideas for the model＇s collision avoidance, in order to improve the reliability of the local path planning. Simulation and sea test results show that this algorithm is a better solution for collision avoidance of the self navigation model under the interference of sea winds and waves with good adaptability.
基金the National Natural Science Foundation of China(Grant Nos.12072262 and 11672231).
文摘It is shown that we can control spatiotemporal chaos in the Frenkel-Kontorova(FK)model by a model-free control method based on reinforcement learning.The method uses Q-learning to find optimal control strategies based on the reward feedback from the environment that maximizes its performance.The optimal control strategies are recorded in a Q-table and then employed to implement controllers.The advantage of the method is that it does not require an explicit knowledge of the system,target states,and unstable periodic orbits.All that we need is the parameters that we are trying to control and an unknown simulation model that represents the interactive environment.To control the FK model,we employ the perturbation policy on two different kinds of parameters,i.e.,the pendulum lengths and the phase angles.We show that both of the two perturbation techniques,i.e.,changing the lengths and changing their phase angles,can suppress chaos in the system and make it create the periodic patterns.The form of patterns depends on the initial values of the angular displacements and velocities.In particular,we show that the pinning control strategy,which only changes a small number of lengths or phase angles,can be put into effect.
