Learned Indexing for Similarity Searching

当面临存储和检索复杂、非结构化或高维数据的任务时（例如，多媒体数据），度量空间通常被用作它们的组织的基础数学概念。因此，可以用来排列数据的唯一度量是数据对象之间的成对相似性。相似性搜索是指用于管理数据的一系列方法，这些方法可以在此类空间中进行有效搜索。相似性搜索的主要范式几十年来基本保持不变-数据对象根据它们的相互距离组织成层次结构，使用代表性枢轴来减少有效搜索数据所需的距离计算数量。我们计划研究这种范式的替代方案，使用机器学习模型来代替枢轴，从而将相似性搜索作为分类问题。我们将使用监督和非监督的方法来实现我们的解决方案。我们还将解决可伸缩性和动态性的问题，并验证度量数据的应用程序。

When faced with the task of storing and retrieving complex, unstructured or high-dimensional data (e.g., multimedia data), metric spaces are often employed as an underlying mathematical concept for their organization. Consequently, the only measure that can be used to arrange the data is a pairwise similarity between data objects. Similarity searching refers to a range of methods used to manage data enabling efficient search in such spaces. The main paradigm of similarity searching has remained mostly unchanged for decades - data objects are organized into a hierarchical structure according to their mutual distances, using representative pivots to reduce the number of distance computations needed to efficiently search the data. We plan to investigate an alternative to this paradigm, using machine learning models to replace pivots, thus, posing similarity search as a classification problem. We will use both supervised and unsupervised approaches to implement our solutions. We will also address the questions of scalability and dynamicity, and verify the applications for metric data.