基于行人属性先验分布的行人再识别

为了提高基于深度学习和属性学习的行人再识别的识别精度,提出一种联合识别行人属性和行人ID的神经网络模型.相对于已有的同类方法,该模型有三大优点:1)为了提高网络在微调后的判别能力,在网络中增加了一层保证模型迁移能力的全连接层; 2)基于各属性样本的数量,在损失函数中对各属性的损失进行了归一化处理,避免数据集中属性类之间的数量不均衡对识别效果的影响; 3)利用数据中各属性分布的先验知识,通过数量占比来调整各属性在损失层中的权重,避免数据集中各属性正负样本的数量不均衡对识别的影响.实验结果表明,本文提出的算法具有较高的识别率,其中在Market 1501数据集上,首位匹配率达到了86.90%,在DukeMTMC数据集上,首位匹配率达到了72.83%,在PETA数据集上,首位匹配率达到了75.68%,且对光照变化、行人姿态变化、视角变化和遮挡都具有很好的鲁棒性.

Bayesian Prediction of Performance in Clinical Trials

This paper presents one of many possible applications of Bayesian inference predictive context of planned tests. We are particularly interested in the use of predictive Bayesian approach in clinical trials or objective is the development of important evidence of an effect of interest We offer the procedure based on the notion of satisfaction index which is a function of the p-value and we look forward, given the available data to calculate a forecast for future satisfaction data as predictive Bayesian hope this index conditional on previous observations. To illustrate the proposed procedure, several models have been studied by choosing the prior distribution justify the reasons of objectivity or neutrality that underlie the analysis of experimental data.