Newman有理插值算子的一个扩充

构造了一个新的有理插值算子 ,它是 Newman算子的一个扩充 ,并且建立了在 [-1 ,1 ]上用这个算子逼近 |x|的渐近估计 .     

A knowledge component extraction technology using figures and tables

Abstract

We show that certain very different types of tasks can be modelled by one rule of abduction which we call coh(2), informally described as being able to go from the conclusions to the premises if the premises cohere, i.e. there is some semantic link between them. These tasks might come from the areas of natural language understanding, planning, perception, etc. Furthermore, there is a family of very closely related rules of coherence coh(n) that can do in one step what otherwise could require iteration of coh(2). More specifically, we formalize proximity-based reasoning as a kind of abduction; we show how it can be captured by such rules; and although we first present these rules informally, we later prove that their meaning can be made precise, and that it is possible to establish some formal properties of them. Of course, this requires a formalism for abduction, which is also sketched in the paper. All this leads to some mathematical results to the effect that certain natural strategies for using rules of coherence are sound and some others are not. Finally, we list some open problems that might be solvable by this approach.     

基于神经网络的摩擦学设计知识表达方法

本文以神经网络理论为基础，提出了摩擦学设计知识表达理论。结合摩擦学设计知识的特点，研究了基于神经网络知识表达方法的实现技术，并给出具体表达实例。实践证明这种方法是行之有效的。     

基于骨骼约束的人体运动捕捉数据失真恢复

针对人体运动捕捉(Motion?Capture, MOCAP)数据实际采集过程中, 由于光线等因素影响而可能出现的同一帧中相邻标记点在时间域上连续缺失的情形, 利用MOCAP数据中存在的潜在相关性和同一运动序列中人体骨骼长度不变特性, 提出一种新的MOCAP数据失真恢复算法. 该算法首先对MOCAP数据进行预处理, 使变换后的数据表示的是相邻标记点的相对位置的变化, 由此得到人体骨骼长度约束项, 再利用稀疏表示和人体骨骼长度约束项进行字典训练, 最后利用训练得到的字典对缺失的数据进行恢复. 通过实验对比表明该算法在提高缺失点坐标恢复精度的同时, 将骨骼长度恢复精度提高到10^–4 cm, 验证了算法的可行性和有效性.