The Keogh Lab

1

The Keogh Lab

Data Mining and Structure Retrieval

Presented byAbdullah Mueen

2

Overview of our work• Our Goal: Extract information from raw, noisy, massive,

unstructured data.• We develop algorithms for

– Classification– Clustering– Rule finding– Motif discovery– Discord discovery– Shapelet discovery– Linkage discovery

• We work closely with the domain experts. – For collecting new data.– To verify our results.

3

Case 1: Motif DiscoveryBeet Leafhopper (Circulifer tenellus)

plant membrane

Stylet

voltage source

input resistor

V

0 50 100 150 2000

10

20

to insectconductive glue

voltage reading

to soil near plant

Exact Discovery of Time Series Motifs.Abdullah Mueen, Eamonn Keogh, Qiang Zhu, Sydney

Cash,  Brandon Westover. SDM 2009.

MK motif discovery

4

false nettles

stinging nettles

Case 2: Shapelet Discovery

false nettles

Shapelet

stinging nettles

Time Series Shapelets: A New Primitive for Data Mining.

Lexiang Ye and Eamonn Keogh. SIGKDD 2009 

5

Case 3: Linkage Discovery

CK-1

0.6291

CK-1

0.9033

CK-1 Distance Measure

0.6

0.7

0.8

0.9

CK-1

Dist

ance Single Linkage Dendrogram

Print House 1 Print House 2

A Compression Based Distance Measure for Texture. Bilson Campana and Eamonn Keogh . SDM 2010

text

a hand-press bookcharacter matrix

textornaments text

Lab Members

Dr. Eamonn KeoghDr. Gustavo Batista

Abdullah MueenQiang Zhu

Bilson CampanaThanawin Art R.

Bing HuYuan Hao

Jesin Zakaria6

7

Motif in Online Data • Maintain motif in streaming data without

introducing latency.

8

Motion Motif• Find repeated motion in motion capture data

which is a 32 dimensional time series.