Learned Spatial Indexes

MPhil Thesis Defence


Title: "Learned Spatial Indexes"

By

Mr. Panagiotis SIMATIS


Abstract

Machine Learning has a plethora of applications in computer science and 
every-day life, spanning from image processing to video game AI. The usage 
of machine learning models, such as Artificial Neural Networks, has 
yielded significant benefits in the execution of tasks, that would 
otherwise be impractical via conventional algorithms. In this thesis, we 
apply machine learning to Spatial Indexing to develop Learned Spatial 
Indexes (LSI). An index is a common tool of database systems, that enables 
speedy retrieval of information. The goal is to construct an index based 
on neural networks, which learns the locations of objects in space, so 
that the LSI promptly returns the desired data given a location-based 
query. We developed several LSIs able to handle the common functions of a 
spatial index, i.e., spatial query processing and data updates. In 
addition, we compared our proposed methods to state of the art 
conventional counterparts. Extensive performance evaluation verifies that 
the LSI is highly efficient, and an exciting topic worth researching 
further.


Date:  			Monday, 24 August 2020

Time:			2:00pm - 4:00pm

Zoom meeting:		https://hkust.zoom.us/j/9273761473

Committee Members:	Prof. Dimitris Papadias (Supervisor)
 			Prof. Raymond Wong (Chairperson)
 			Dr. Pan Hui


**** ALL are Welcome ****