Learning bipartite Bayesian networks under monotonicity restrictions

ABSTRACT

Learning parameters of a probabilistic model is a necessary step in machine learning tasks. We present a method to improve learning from small datasets by using monotonicity conditions. Monotonicity simplifies the learning and it is often required by users. We present an algorithm for Bayesian Networks parameter learning. The algorithm and monotonicity conditions are described, and it is shown that with the monotonicity conditions we can better fit underlying data. Our algorithm is tested on artificial and empiric datasets. We use different methods satisfying monotonicity conditions: the proposed gradient descent, isotonic regression EM, and non-linear optimization. We also provide results of unrestricted EM and gradient descent methods. Learned models are compared with respect to their ability to fit data in terms of log-likelihood and their fit of parameters of the generating model. Our proposed method outperforms other methods for small sets, and provides better or comparable results for larger sets.     

宽带毫米波大规模MIMO系统中低复杂度波束域信道估计

针对宽带毫米波大规模MIMO系统信道估计精度低及实现复杂度较高的问题，在传统支撑检测方案的基础上提出一种基于Gauss-Seidel方法的串行支撑检测（GS-SSD）方案。该方案不使用共同支撑假设，参考串行干扰删除，将整体信道估计问题分解为一系列子问题，每个子问题仅考虑一个信道成分。同时，利用Gauss-Seidel方法近似高复杂度的矩阵求逆。仿真结果表明，相比于基于串行支撑检测（SSD）的方案， GS-SSD方案在将求逆复乘数降低一个数量级的同时可以取得接近SSD方案的信道估计性能。     

富水软岩巷道稳定性控制技术研究

富水条件下软岩巷道容易出现大变形及围岩强度大幅度降低等情况,对矿井的生产安全造成极大威胁。为探索富水软岩巷道稳定性控制技术,结合新上海一号煤矿113082工作面运输巷工程实例,采用理论分析方法研究了地下水对巷道围岩及支护结构的影响,提出围岩稳定性控制对策,并对控制方案进行数值模拟验证。针对富水软岩巷道,首先采取导、疏结合的控水措施,减小对原有围岩强度的影响,降低对支护结构的损害;其次采用全断面、多手段联合加固的支护措施,控制围岩变形。工程实际及数值模拟结果表明,对于富水软岩巷道采取以上技术措施,能够有效保证巷道的稳定性。     

谈淤泥质复杂基坑地基处理及支护技术

淤泥质地形在施工过程中可能因地基强度不够而发生事故,施工难度高,对此类地基处理及支护技术的研究较为重要。     

Alcoholism detection based on Hu moment invariants of event-related potentials

Diagnostic efforts to identify alcoholism are frequently subjective; however, recent technological advances have allowed the automated assessment of objective signals. To improve the performance of the automated classification of patients with alcohol dependence (AD) and controls, we analyzed features of event-related potentials (ERPs) in 72 subjects with AD and 42 controls while the subjects underwent three different visual oddball tasks. In addition to such commonly used morphological features (MFs) as time-domain amplitude and latency components as well as discrete wavelet transform coefficient entropies (WEs), we analyzed Hu moment invariants (HMIs) in this study. Smaller P3 and P2 amplitudes, longer latencies, lower WE values, and lower HMI values were observed for AD subjects relative to those observed for controls. We used a Naïve Bayes algorithm to compare classification accuracies among MFs, WEs, HMIs, combinations of MFs/WEs, and combinations of all three characteristics. We found that HMIs improved the classification accuracy by 17%. Our findings demonstrated that a combined feature analysis of ERPs obtained during a visual oddball task can be a useful and discriminative tool for diagnostic purposes in alcoholism and that HMIs show promise for improving the objective diagnosis of alcoholism and similar diseases.     

A deep neural network based classifier for brain tumor diagnosis

Classification process plays a key role in diagnosing brain tumors. Earlier research works are intended for identifying brain tumors using different classification techniques. However, the False Alarm Rates (FARs) of existing classification techniques are high. To improve the early-stage brain tumor diagnosis via classification the Weighted Correlation Feature Selection Based Iterative Bayesian Multivariate Deep Neural Learning (WCFS-IBMDNL) technique is proposed in this work. The WCFS-IBMDNL algorithm considers medical dataset for classifying the brain tumor diagnosis at an early stage. At first, the WCFS-IBMDNL technique performs Weighted Correlation-Based Feature Selection (WC-FS) by selecting subsets of medical features that are relevant for classification of brain tumors. After completing the feature selection process, the WCFS-IBMDNL technique uses Iterative Bayesian Multivariate Deep Neural Network (IBMDNN) classifier for reducing the misclassification error rate of brain tumor identification. The WCFS-IBMDNL technique was evaluated in JAVA language using Disease Diagnosis Rate (DDR), Disease Diagnosis Time (DDT), and FAR parameter through the epileptic seizure recognition dataset.     

结合深度学习与支持向量机的金属零件识别

目的 在视觉引导的工业机器人自动拾取研究中，关键技术难点之一是机器人抓取目标区域的识别问题。特别是金属零件，其表面的反光、随意摆放时相互遮挡等非结构化因素都给抓取区域的识别带来巨大的挑战。因此，本文提出一种结合深度学习和支持向量机的抓取区域识别方法。方法 分别提取抓取区域的方向梯度直方图（HOG）和局部二进制模式（LBP）特征，利用主成分分析法（PCA）对融合后的特征进行降维，以此来训练支持向量机（SVM）分类器。通过训练Mask R-CNN（regions with convolutional neural network）神经网络完成抓取区域的初步分割。然后利用SVM对Mask R-CNN识别的抓取区域进行二次分类，完成对干扰区域的剔除。最后计算掩码完成实例分割，以此达到对抓取区域的精确识别。结果 对于随机摆放的铜质金属零件，本文算法与单一的Mask R-CNN及多特征融合的SVM算法就识别准确率、错检率、漏检率3个指标进行了比较，结果表明本文算法在识别准确率上较Mask R-CNN和SVM算法分别提高了7%和25%，同时有效降低了错检率与漏检率。结论 本文算法结合了Mask R-CNN与SVM两种方法，对于反光和遮挡情况具有一定的鲁棒性，同时有效地提升了目标识别的准确率。     

基于需求预测的PaaS平台资源分配方法

针对缺乏PaaS平台下资源需求的有效预测与优化分配的问题，提出一种资源需求预测模型和分配方法。首先，根据PaaS平台中应用对资源需求的周期性来对资源序列进行切分，并在短期预测的基础上结合应用的多周期性特征，利用多元回归算法建立综合的预测模型。然后，基于MapReduce架构设计实现了一个Master-Slave模式的PaaS平台资源分配系统。最后，结合当前任务请求和资源需求预测结果进行资源分配。实验结果表明，采用该资源需求预测模型和分配方法后，相比于自回归模型和指数平滑算法，平均绝对百分比误差分别下降8.71个百分点和2.07个百分点，均方根误差分别下降2.01个百分点和0.46个百分点。所提预测模型的预测结果不仅误差小，与真实值的拟合程度也较高，而且利用较小的时间开销就可以获得较高的准确度。此外，使用该预测模型的PaaS平台的资源请求的平均等待时间有了明显的下降。     

Looking into the TESSERACT: Time-drifts in event streams using series of evolving rolling averages of completion times

Business processes are dynamic and change due to diverse factors. While existing approaches aim to detect drifts in the process structure, Tesseract looks for temporal drifts in activity interim times. This orthogonal view on the process extends the traditional data cube of events – case id, activities and timestamps – by a fourth dimension and improves the operational support by a visualization of temporal drifts in real-time.Insights about temporal deviations lead to an augmented awareness of imminent failures or improved service times. The detection of related structural concept drifts can be improved by early warning, as operation times of critical parts often increase before they catastrophically fail.     

The application of decision support tools and the influence of local data in prioritizing barrier removal in lower Michigan,USA

Web-based decision support tools (DSTs) can be useful to facilitate decision-making processes for managing complex natural resource systems. However, the alignment of DSTs with the objectives in governmental policies or management plans and the influence of limited local data on the outputs of these tools may reduce the use of DSTs by decision makers. In this study, we examined the outcomes of web-based DSTs when different types of local data were incorporated and demonstrated a way to incorporate outputs from multiple DSTs or local inventories to benefit barrier removal decisions. Restoring habitat connectivity in rivers in northwest lower Michigan, USA, was used as a case study due to the abundance of local inventory data and web-based DSTs. We found that, when compared to prioritizations made using local data, some DSTs could produce similar outcomes (in barriers selected, cost, and the benefit for migratory fish) with limited data, but the trade-offs among users' objectives might influence the cost and effectiveness of DSTs' outputs. Improving the ability of DSTs to incorporate objectives consistent with policy and stakeholders' values (e.g., restore certain species or sedimentation control) across management scales can help close the gap between tool recommendations and management decisions while making the barrier removal prioritization process transparent and efficient.