Historical Learned Index

MPhil Thesis Defence


Title: "Historical Learned Index"

By

Mr. Kai WANG


Abstract

Conventional database indexes are ephemeral because they manage data at 
the current time. On the other hand, many applications require the 
maintenance of old data versions. For instance, in most countries the 
history of bank accounts must be kept over a period of years to satisfy 
state regulations. When the balance of an account changes (e.g., due to 
deposits or withdrawals), a new version of the record is created and the 
old one is saved, together with its lifespan (i.e., the time interval 
during which the balance had the previous value). Ephemeral structures can 
be converted to historical indexes, using the multi-version framework. In 
recent years, the application of machine learning techniques, especially 
linear models, on ephemeral indexes has shown significant promise. The 
thesis integrates linear models and the multi-version framework, to 
develop the first learned historical index. We propose detailed update 
algorithms and perform extensive experiments to investigate the 
feasibility and effectiveness of the novel index.


Date:  			Tuesday, 31 January 2023

Time:			11:00am - 1:00pm

Venue:			Room 3494
 			lifts 25/26

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


**** ALL are Welcome ****