• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>Machine learning >Speeding-up Nearest Neighbour Memories: The Template Tree Case Memory Organisation
【24h】

Speeding-up Nearest Neighbour Memories: The Template Tree Case Memory Organisation

机译：加速最近的邻居内存：模板树案例存储组织

获取原文
获取原文并翻译 | 示例
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Case retrieval is a crucial part of any CBR system as it provides the raw material for constructive solution generation by the reasoner. Case memories, as substrates for retrieval, are based on one of two main techniques: nearest neighbours classification or common feature indexing. While the former admits the incorporation of sophisticated similarity functions for accurate case retrieval, the latter provides rapid access to the stored cases. In this paper, we propose a case memory based on template trees. For this reason, we introduce the dynamic template trees, which are an adaptation of the classic template trees, originally introduced for classification in the pattern recognition field. They combine the benefits of both major approaches to case memory organisation. The hierarchical organisation of cases speeds-up case acquisition as well as case retrieval. Moreover, as they rely on the integration of an explicit similarity function, they maintain the advantages of nearest neighbour memories. Further, they can be iteratively. Results from experimental evaluation in case-based optimisation of the NP-hard capacitated facility location problem show that the proposed memory organisation exhibits a considerably better scale-up behaviour than does a conventional flat list memory. This scale-up is achieved without degrading solution quality.
机译：案例检索是任何CBR系统的关键部分，因为它为推理者提供了生成建设性解决方案的原材料。案例存储作为检索的基础，基于两种主要技术之一：最近邻居分类或公共特征索引。前者承认结合了复杂的相似功能以进行准确的案件检索，而后者则提供了对存储案件的快速访问。在本文中，我们提出了一种基于模板树的案例存储。因此，我们介绍了动态模板树，它们是对经典模板树的改编，最初是为模式识别领域中的分类而引入的。它们结合了两种主要的案例存储组织方法的优点。案例的层次结构加快了案例获取和案例检索的速度。此外，由于它们依靠显式相似性函数的集成，因此它们保持了最近邻居存储器的优势。此外，它们可以迭代。在基于案例的NP硬能力设施位置问题的优化中的实验评估结果表明，与传统的平面列表存储器相比，拟议中的存储器组织具有更好的按比例扩展行为。在不降低解决方案质量的情况下实现了这种放大。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号