Database systems have been the critical part of every data-intensive
application in the past decades. Nowadays machine learning techniques
open up the opportunities to improve the performance as well as
usability of database systems by replacing some human labored tasks or
heuristic approached with learned models. In this seminar, I will
introduce two pieces of up-to-date research work in the database
community. The first one is learned index structure. The basic idea is
of the first paper is to regard index structures in the database as
mapping a key to the position of a record within an array. Then the
functionality of indexes can be realized by several trained models.
The second one is query-based workload forecasting for a database
system. To make the database system “self-driving”, it requires the
database system to automatically choose its optimizations without
human intervention. To reach this goal, the second paper constructs a
toolkit QB5000, which can predict the incoming workload of DBMS by
first clustering query workloads into templates and then use a variety
of machine learning models, such as LR, KR, and RNN for forecasting.

References:

[1] Tim Kraska, Alex Beutel, Ed H. Chi, Jeffrey Dean, Neoklis
Polyzotis. The Case for Learned Index Structures. SIGMOD Conference
2018: 489-504.

[2] Lin Ma, Dana Van Aken, Ahmed Hefny, Gustavo Mezerhane, Andrew
Pavlo, Geoffrey J. Gordon.Query-based Workload Forecasting for
Self-Driving Database Management Systems. SIGMOD Conference 2018:
631-645.