机器学习在量子通信资源优化配置中的应用

在未来量子通信网络的大规模应用中,如何根据当前用户实际情况实现资源优化配置,比如选择最优量子密钥分发协议(quantum key distribution, QKD)和最优系统参数等,是实现网络应用的一个重要考察指标.传统的QKD最优协议选择以及参数优化配置方法,大多是通过局部搜索算法来实现.该方法需要花费大量的计算资源和时间.为此,本文提出了将机器学习算法应用到QKD资源优化配置之中,通过回归机器学习的方式来同时进行不同情境下的最优协议选择以及最优协议的参数优化配置.此外,将包括随机森林(random forest, RF)、最近邻(k-nearest neighbor, KNN)、逻辑回归(logistic regression)等在内的多种回归机器学习模型进行对比分析.数据仿真结果显示,基于机器学习的新方案与基于局部搜索算法的传统方案相比,在资源损耗方面实现了质的跨越,而且RF在多个回归评估指标上都取得了最佳的效果.此外,通过残差分析,发现以RF回归模型为代表的机器学习方案在最优协议选择以及参数优化配置方面具有很好的环境鲁棒性.因此,本工作将对未来量子通信网络实际应用起到重要的推进作用.

商业银行条件风险价值的分位数回归组合估算

基于线性分位数回归和三种非线性分位数回归模型,对我国14家上市商业银行的股价收益率VaR与风险状态下银行的系统风险CoVaR进行了估算。通过6种回测法的8个检验统计量对模型进行了综合评估,并基于单一模型构建了更加综合的组合模型。研究表明:就准确性和稳健性而言,不同机器学习分位数回归模型各有所长;线性分位数回归模型较综合,可应用情景多;由线性分位数回归、神经网络分位数回归和随机森林分位数回归构建的组合模型比单一模型更加准确有效,综合有效性最强。

基于决策树模型的黄河水沙变化预测

本文基于小浪底水库下游黄河某水文站2016—2021年的水流量与含沙量的实际监测数据,分别建立随机森林(Random Forest)、决策树(Decision Tree)和极端梯度提升(XGBoost)3种机器学习回归模型预测水流量和含沙量的走势,并对比3种模型的拟合效果。结果表明,与随机森林和极端梯度提升算法相比,决策树算法对水沙变化的预测效果更好,其能够有效拟合水沙变化的走势,对未来黄河流域的水沙治理有一定参考价值。

Efficient Stochastic Simulation Algorithm for Chemically Reacting Systems Based on Support Vector Regression

The stochastic simulation algorithm （SSA） accurately depicts spatially homogeneous wellstirred chemically reacting systems with small populations of chemical species and properly represents noise, but it is often abandoned when modeling larger systems because of its computational complexity. In this work, a twin support vector regression based stochastic simulations algorithm （TS^3A） is proposed by combining the twin support vector regression and SSA, the former is a well-known robust regression method in machine learning. Numerical results indicate that this proposed algorithm can be applied to a wide range of chemically reacting systems and obtain significant improvements on efficiency and accuracy with fewer simulating runs over the existing methods.

Machine learning-based prediction of soil compression modulus with application of ID settlement

The compression modulus(Es)is one of the most significant soil parameters that affects the compressive deformation of geotechnical systems,such as foundations.However,it is difficult and sometime costly to obtain this parameter in engineering practice.In this study,we aimed to develop a non-parametric ensemble artificial intelligence(AI)approach to calculate the Es of soft clay in contrast to the traditional regression models proposed in previous studies.A gradient boosted regression tree(GBRT)algorithm was used to discern the non-linear pattern between input variables and the target response,while a genetic algorithm(GA)was adopted for tuning the GBRT model's hyper-parameters.The model was tested through 10-fold cross validation.A dataset of 221 samples from 65 engineering survey reports from Shanghai infrastructure projects was constructed to evaluate the accuracy of the new model5 s predictions.The mean squared error and correlation coefficient of the optimum GBRT model applied to the testing set were 0.13 and 0.91,respectively,indicating that the proposed machine learning(ML)model has great potential to improve the prediction of Es for soft clay.A comparison of the performance of empirical formulas and the proposed ML method for predicting foundation settlement indicated the rationality of the proposed ML model and its applicability to the compressive deformation of geotechnical systems.This model,however,cannot be directly applied to the prediction of Es in other sites due to its site specificity.This problem can be solved by retraining the model using local data.This study provides a useful reference for future multi-parameter prediction of soil behavior.

Affective rating ranking based on face images in arousal-valence dimensional space

In dimensional affect recognition, the machine learning methods, which are used to model and predict affect, are mostly classification and regression. However, the annotation in the dimensional affect space usually takes the form of a continuous real value which has an ordinal property. The aforementioned methods do not focus on taking advantage of this important information. Therefore, we propose an affective rating ranking framework for affect recognition based on face images in the valence and arousal dimensional space. Our approach can appropriately use the ordinal information among affective ratings which are generated by discretizing continuous annotations.Specifically, we first train a series of basic cost-sensitive binary classifiers, each of which uses all samples relabeled according to the comparison results between corresponding ratings and a given rank of a binary classifier. We obtain the final affective ratings by aggregating the outputs of binary classifiers. By comparing the experimental results with the baseline and deep learning based classification and regression methods on the benchmarking database of the AVEC 2015 Challenge and the selected subset of SEMAINE database, we find that our ordinal ranking method is effective in both arousal and valence dimensions.

Prediction for Geometric Characteristics of Single Track of Deposition Layer and Surface Roughness in Thin Wire-Based Metal Additive Manufacturing Process

Machine learning prediction models for thin wire-based metal additive manufacturing(MAM)process were proposed,aiming at the complex relationship between the process parameters and the geometric characteristics of single track of the deposition layer and surface roughness.The effects of laser power,wire feeding speed and scanning speed on the width and height of the single track and surface roughness were experimentally studied.The results show that laser power has a significant impact on the width of the single track but little effect on the height.As the wire feeding speed increases,the width and height of the single track increase,especially the height.The faster the scanning speed,the smaller the width of the single track,while the height does not change much.Then,support vector regression(SVR)and artificial neural network(ANN)regression methods were employed to set up prediction models.The SVR and ANN regression models perform well in predicting the width,with a smaller root mean square error and a higher correlation coefficient R2.Compared with the ANN model,the SVR model performs better both in predicting geometric characteristics of single track and surface roughness.Multi-layer thin-walled parts were manufactured to verify the accuracy of the models.