基于单光子计数激光雷达的Bayesian检测研究

在单光子计数激光雷达检测领域,目前的检测方法在低信噪比情况下虚警概率会增加,同时也无法适应噪声变化的问题。针对这些问题,提出了一种基于Bayesian的检测方法,该方法首先通过雷达方程估计回波信号光子数的范围,将其作为先验信息,而后结合二项分布建立了累计概率模型,基于Bayesian判决准则计算得到检测阈值,此阈值能够在检测概率与虚警概率中间择其平衡。这种方法不仅克服了低信噪比检测困难的情况,还减少了先验信息的获取难度。实验结果表明,对比固定阈值其虚警概率降低了10倍。对比“恒虚警”其检测概率提高了约20。验证了方法具有良好的检测效果,具备一定的可操作性。     

提升管内颗粒浓度梯度力的特性

由于Kutta-Joukowski横向力与浓度梯度力的共同作用，提升管内颗粒沿径向在边壁大量聚集并形成稳定的环?核结构。根据实验数据，分析了颗粒浓度梯度的径向分布特征，考察了不同操作条件下浓度梯度力系数K的分布特性。由Kutta-Joukowski横向力与浓度梯度的关系，提出了浓度梯度力的表达式Fρ=K(dρ/dr)A及浓度梯度力系数K的表达式K=[?ρg(νg?νp)(?v/?r)r]/?dρ/dr+(d2ρ/dr2)r?。提升管内颗粒群受到的浓度梯度力与浓度梯度力系数K有关。浓度梯度力系数K在提升管中心处为0，沿提升管径向呈“N”型分布，随表观气速增加而增加；提升管内充分发展段K的数值明显大于提升管加速区和出口约束区，总结了浓度梯度力系数K的经验关联式。     

Adaptive Discretization for Computerized Tomography

Two adaptive discretization frameworks are tested for computerized tomography (CT) data reconstruction. Removal of inactive pixels is primary motivation. Efficient and user independent entropy optimized masking is employed for spatial filtering purposes. Density of nodes at high gradient of reconstructed physical property is used as adaptation criterion. An alternative option, independent from noisy projection data and nature of the physical properties, is also discussed. Sensitivity analysis between the uniform and nonuniform (evolved via adaptive route) reconstruction grid reveals the utility of nonuniform grids. Iterative and transform based reconstruction techniques are used. Outcomes are tested successfully on three real world projection data from two different compact CT setups and one commercial high-resolution micro-CT scanner.     

基于多尺度的贝叶斯模型显著性检测

针对传统基于贝叶斯模型的显著性检测算法存在准确率不理想的问题,提出了一种基于多尺度的贝叶斯模型显著性检测算法。通过超像素分割算法(SLIC)将原图分割成不同尺度的超像素,根据超像素边界信息得到背景种子,进而通过距离计算和多尺度融合得到背景先验;对原图进行颜色增强,采用Harris算子对增强图进行检测角点求得凸包,融合不同尺度下的超像素得到凸包先验;融合背景先验和凸包先验得到最终先验;利用颜色直方图和凸包计算似然概率;将最终先验和似然概率通过贝叶斯模型计算显著图。在公开数据集MSRA1000、ECSSD上与多种传统算法进行准确率和召回率对比,该算法有更好的表现。     

贝叶斯网络查询语义扩展的专家发现方法

专家发现是实体检索领域的一个研究热点，针对经典专家发现模型存在索引术语独立性假设与检索性能低的缺陷，提出一种基于贝叶斯网络模型的专家发现方法。该方法模型采用四层网络结构，能够实现图形化的概率推理，同时运用词向量技术能够实现查询术语的语义扩展。实验结果显示该模型在多个评价指标上均优于经典专家发现模型，能够有效实现查询术语语义扩展，提高专家检索性能。     

Mathematical formulation and heuristic algorithms for optimisation of auto-part milk-run logistics network considering forward and reverse flow of pallets

The operational planning of distribution network for automotive industry is complex with many conditions to consider, including heterogeneous fleet, enforcing the feasibility of 3D-packing of pallets into vehicles to address the vehicle's capacity in terms of weight and volume, compatibility of orders in a vehicle, returning empty pallets from assembly-plants backwards to suppliers, and delivery time windows. A mathematical model (MILP) is proposed that takes account of these conditions to minimise total transportation costs. The network structure can be a combination of direct shipment and milk-run for both forward and reverse flow of pallets. The model is solved optimally for small-size problems. For solving larger problems, a heuristic algorithm (in two versions) is proposed that uses a similarity measure to generate a reasonable list of orders. Best/first-fit strategies are employed to generate a feasible solution with the aid of a relaxed version of the proposed MILP. Improvement heuristics are also designed. Unlike most of existing constructive heuristics, our aim for developing the heuristic approach is to force routing decision, with all of its considerations, being made optimal. We also use the proposed best-fit strategy in the body of grouping evolution strategy (GES) algorithm to attain an effective meta-heuristic approach. The effectiveness of heuristics is tested on generated instances which demonstrates they are optimal for small-size problems. They are also tested on the data of daily auto-parts shipments gathered from the largest Iranian automobile company. Results demonstrate there exists a significant potential for cost saving through milk-run strategy compared with the direct shipping strategy.     

Formulation strategy towards minimizing viscosity mediated negative food effect on disintegration and dissolution of immediate release tablets

Food induced viscosity can delay disintegration and subsequent release of API from solid dosage form which may lead to severe reduction in the bioavailability of BCS type III compounds. Formulations of such tablets need to be optimized in view of this postprandial viscosity factor. In this study, three super disintegrants, croscarmellose sodium (CCS), cross-linked polyvinylpolypyrrolidone (CPD), and sodium starch glycolate (SSG) were assessed for their efficiency under simulated fed state. Tablets containing these disintegrants were compressed at 10 and 30?KN, while taking lactose as a soluble filler. In addition to other compendial tests, disintegration force of these formulations was measured by texture analysis. Comparison of parameters derived from force – time curves revealed a direct relation of maximum disintegration force (F_max) and disintegration force development rate (DFDR) with compressional force in fasted state, whereas an inverse relationship of F_max and DFDR with compressional force was observed in fed state. The gelling tendency of disintegrants influenced the rate of release of API in simulated fed and fasted states when compressional force was changed. These observations recommend the evaluation of formulations in simulated fed state, in the development stage, with an objective of minimizing the negative impact of food induced viscosity on disintegration. Use of disintegrants that act without gelling or can counteract the effect of gelling is recommended for tablet formulations with reduced disintegration time (DT) and mean dissolution time (MDT) in fed state, respectively.     

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.     

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.     

Formulating Pantone colors by unused base inks,formulation software,and a spectrophotometer

Unused base inks that are not going to be used for printing production are considered to be hazardous materials. Their disposal is expensive, and strict environmental regulations should be followed for their disposal. As an alternative, this article describes how spectral data of unused base inks can be gathered and mixed to generate new colors to incorporate them back to print production for small‐volume jobs. In this study, 30 different Pantone colors were selected as target colors. The CIE L*a*b* spectral data of Pantone colors and unused base inks were gathered via a spectrophotometer. A commercial formulation software, based on multiflux theory and CIE L*a*b* color space, was used to formulate ink recipes that contained the base inks. To quantify the performance of ink recipes, they were mixed and printed using an offset printability tester. The CIELAB ΔE*_ab metric, developed by CIE, was used to detect the visual differences between the target Pantone Color and printed colors.