Full Citation

Title: NeuroDB: A Generic Neural Network Framework for Efficiently and Approximately Answering Range-Aggregate and Distance to Nearest Neighbor Queries

Citation Type: Miscellaneous

Publication Year: 2020

ISBN:

ISSN:

DOI:

NSFID:

PMCID:

PMID:

Abstract: We observe that database queries can be considered as functions that can be approximated. This allows us to formulate a learning task to learn a single model to answer any query. This is different than many recent studies that learn a model for a specific query type, since we can learn the model for any query type. We formalize this observation, formulate the problem of learning to answer database queries and introduce NeuroDB, a generic neural network framework that can learn to answer different query types approximately. As a proof of concept, we show that NeuroDB can be specifically used to answer distance to nearest neighbour query and range aggregate queries, two important building blocks of many real-world applications. We experimentally show that NeuroDB answers these two query types with orders of magnitude improvement in query time over the state-of-the-art competitions, and by constructing a model that takes only a fraction of data size. NeuroDB achieves this by learning existing patterns between query input and output and by exploiting the query and data distributions. Furthermore, the same neural network architecture is used to answer both query types, bringing to light the possibility of using a generic framework to answer different unrelated query types efficiently.

Url: https://infolab.usc.edu/DocsDemos/NeuroDB.pdf

User Submitted?: No

Authors: Zeighami, Sepanta; Shahabi, Cyrus

Publisher:

Data Collections: IPUMS USA

Topics: Housing and Segregation

Countries: