A Comparison between Structural and Embedding Methods for Graph Classification
A Comparison between Structural and Embedding Methods for Graph Classification
复制标题
图分类的结构方法和嵌入方法的比较
DOI：
10.1007/978-3-642-34166-3_26
复制
发表时间：
2012
期刊：
影响因子：
--
通讯作者：
F. Serratosa
中科院分区：
文献类型：
--
作者：
Albert Solé;X. Cortés;F. Serratosa
结构模式识别是一个众所周知的研究领域，诞生于20世纪80年代初。 30年来，诸如图之类的结构都是通过直接使用节点和弧上的属性值的函数优化来进行比较的。然而，在过去的十年中，出现了内核和嵌入方法。这些新方法通过将图表示为多维空间来推导节点之间的相似度值和最终标记。相对于经典的结构方法，最近似乎更喜欢核方法和嵌入方法。然而，这两种方法都有优点和缺点。在这项工作中，我们将结构方法与嵌入方法和核方法进行比较。结果表明，在评估的数据集上，一些结构方法的性能稍好，因此，放弃用于图模式识别的经典结构方法还为时过早。
Structural pattern recognition is a well-know research field that has its birth in the early 80s. Throughout 30 years, structures such as graphs have been compared through optimization of functions that directly use attribute values on nodes and arcs. Nevertheless, in the last decade, kernel and embedding methods appeared. These new methods deduct a similarity value and a final labelling between nodes through representing graphs into a multi-dimensional space. It seems that lately kernel and embedding methods are preferred with respect to classical structural methods. However, both approaches have advantages and drawbacks. In this work, we compare structural methods to embedding and kernel methods. Results show that, with the evaluated datasets, some structural methods give slightly better performance and therefore, it is still early to discard classical structural methods for graph pattern recognition.
Ent-Boost ：使用熵度量进行增强以实现鲁棒目标检测，
DOI：
--
发表时间：
2006
期刊：
Proc. of International Conference on Pattern Recognition (ICPR2006)
影响因子：
--
作者：
Duy-Dinh Le;Shin'ichi Satoh
通讯作者：
Shin'ichi Satoh