【摘 要】
:
在半监督分类问题中,某些无标签的数据点很有可能属于某一特定的类,则这些点可被信任,应当被充分地利用去帮助学习正确的分类界面,从而提升图像分类的效果.因此,如何区分出可
论文部分内容阅读
在半监督分类问题中,某些无标签的数据点很有可能属于某一特定的类,则这些点可被信任,应当被充分地利用去帮助学习正确的分类界面,从而提升图像分类的效果.因此,如何区分出可信点在半监督分类问题中显得尤为重要.针对该问题,提出一种联合的半监督学习方法,可以自适应地区分可信点,并且提出了相应的迭代优化分类器和可信点标签矩阵的算法,把分类器的归纳能力和无标签数据的标签直推化能力融合到了一个框架中,不需要额外的步骤去预测无标签数据的标签.在两个人脸数据库中大量实验验证了所提方法的可行性和有效性.
其他文献
通过溶剂热法制备Cd0.01Co0.79Cr0.8 S2粉体.XRD、FESEM、TEM、纳米激光粒度分析表明制备的Cd0.01Co0.79Cr0.8S2是纳米级的粉体,颗粒细小、结晶良好,呈现简单立方结构.Cd0.01
许多有重要价值的实际问题的数学模型为极小极大分布鲁棒优化模型,该类模型常存在的分布是不确定的,基于Hellinger距离散度,探讨了极小极大分布鲁棒优化问题的一个等价形式.
This paper deals with a class of Schr¨odinger-Poisson systems. Under some con-ditions, we prove that there exists a ground state solution of the system. The pr
In this article, we study the complete convergence for weighted sums of widely orthant dependent random variables. By using the exponential probability inequali
In this paper, we investigate non-isothermal one-dimensional model of capillary compressible fluids as derived by M Slemrod(1984) and J E Dunn and J Serrin(1985
In this paper, we show the asymptotic limit for the 3D Boussinesq system with zero viscosity limit or zero diffusivity limit. By the classical energy method, we
In this short paper, we firstly give a short proof of Birkhoff-Kakutani Theorem by Moore metrizable Theorem. Then we prove that G is a topological group if it i
Mixed element formats of any order based on bubble functions for the stationary Stokes problem are derived in triangular and tetrahedral meshes and the converge
本文采用传统的高温固相反应法制备Ba0.85 Ca0.15 Ti0.9 Zr0.1O3( BCTZ)陶瓷靶材,并利用脉冲激光沉积(PLD)法,在生长有SrRuO3底电极的SrTiO3(100)衬底上沉积BCTZ(BCTZ/SRO/ST
为了求解非线性矩阵方程Xs+m∑i=1AHiX-tiAi=Q在0<s≤1,t≥1时的最大Hermitian正定解,运用不动点迭代算法和矩阵理论知识,给出了解存在的条件以及算法的收敛性证明.