Research Overview Kyriakos Mouratidis Assistant Professor School of Information Systems Singapore Management University

Research Overview

Kyriakos Mouratidis

Assistant Professor

School of Information Systems

Singapore Management University

http://www.mysmu.edu/faculty/kyriakos/

2

Spatial Queries

q

p1

p2

p6

p5

p4p3

p7

- Indexing spatial data and query processingE.g., “find the 10 closest restaurants to my location”

- K nearest neighbors (if K=2)

3

Continuous Queries

Continuous re-evaluation as data change. Eg:• “monitor who are the 10 SMU students that are

closest to my location as I walk around”

q

p1

p2

p6

p5

p4p3

p7

4

Continuous Queries• Cont. NN in Euclidean space: SIGMOD’05• Cont. NN in road networks: VLDB’06• Cont. Top-k monitoring: SIGMOD’06

– Eg: "continuously report the 5 most interesting stocks according to my investment criteria”

• Cont. Text queries on document streams: TKDE’11

d1

User 1

User 2

d2... d3

Monitoring Server

1

Top-k2 docs

Sliding Window

Incoming Document Stream

5

Spatial Optimization Queries

• E.g.: At which 10 positions in S’pore should McDonalds open branches so that the average distance between clients and their closest branch is minimized?

• E.g.: Given a coverage radius and a maximum capacity of a Mobile Service Provider’s base stations, find a (dynamic) assignment of mobile phone users to a base station so that the average distance between them is minimal.

Spatial Optimization Example

• A set of wireless routers serve a set of laptops– each router can serve at most 3 laptops concurrently– the signal strength (ie, the QoS) drops with distance

• How can we assign laptops to routers so that we1. Serve the maximum possible number of users, AND2. Minimize the average distance between laptops-routers?

• Assignments by “local” criteria (eg, NN below) would fail!

3-Nearest Neighbor Queries

Spatial Optimization Example

• Optimal Assignment:

• Aim: quickly compute the optimal assignment over large datasets [SIGMOD’08, TODS’10]

Location Privacy

• How could an untrusted server answer your spatial queries without learning your location?

• Example: shortest path query [VLDB’12]

8

Destination

Source

9

Building block: Hardware-aided PIR

• Practical PIR = hardware-aided PIR[Williams & Sion: Usable PIR. NDSS’08]

DatabaseClient

Page requests

Data pages

SCP

LBS

SSL connection

Fetching a disk page: amortized comp. cost O(log2N)i.e., approx. 1 sec for a Gigabyte database

10

Verification in Outsourced Databases:

• Model: Database as a Service

• Data Owner uploads DB to untrusted server

• Server hosts the DB and answers queries from users

• How can users verify that the results to their queries are authentic and complete?

• Examples: text queries [VLDB’08], relational/spatial queries [VLDBJ’09], shortest path queries [ICDE’10]…

Other cool stuff

• TripAdvisor has hotel information, such as: price, value, location, cleanliness, user rating

• Imagine this interface to select top-10 options:

11

Immutable Regions [VLDB’13]

Thank you!

12