Table of Contents Table of Contents
Previous Page  17 / 561 Next Page
Information
Show Menu
Previous Page 17 / 561 Next Page
Page Background

Time Blocks

Sunday - Tuesday

8:00am -9:30am

11:00am – 12:30pm

1:30pm – 3:00pm

4:30pm – 6:00pm

Wednesday

8:00am -9:30am

11:00am – 12:30pm

12:45pm -2:15pm

2:45pm – 4:15pm

4:30pm- 6:00pm

TA01

The day of

the week

Time Block.

Matches the time

blocks shown in the Program

Schedule.

Room number.

Room locations are

also indicated in the listing for each

session.

How to Navigate the

Technical Sessions

There are four primary resources to help you

understand and navigate the Technical Sessions:

• This Technical Session listing, which provides the

most detailed information. The listing is presented

chronologically by day/time, showing each session

and the papers/abstracts/authors within each

session.

• The Author and Session indices provide

cross-reference assistance (pages 518-560).

Quickest Way to Find Your Own Session

Use the Author Index (page 518) — the session code

for your presentation will be shown along with the room

location. You can also refer to the full session listing for

the room location of your session.

The Session Codes

17

Sunday, 8:00AM - 9:30AM

SA01

101A-MCC

Temporal Data Mining and Pattern Discovery

Sponsored: Data Mining

Sponsored Session

Chair: Mustafa Gokce Baydogan, Bogazici University, Bebek, Istanbul,

34342, Turkey,

baydoganmustafa@gmail.com

1 - Discovering Distinct Features Using Deep Learning For

Arrhythmia Detection

Seho Kee, Arizona State University, Tempe, AZ, United States,

skee4@asu.edu

, Phillip Howard, George Runger

Although domain knowledge-based features have been widely adopted in

anomaly detection studies, they still suffer from the limitations of the insufficient

known features or unavailability in practice. To address these problems, we

propose an autoencoder model that is able to discover useful features that identify

anomaly patterns in temporal heartbeat data without assuming any prior

knowledge. The results show that the discovered features obtained from just a

two-dimensional projection layer can effectively distinguish abnormal beats from

normal beats without training on pre-labeled data.

2 - Process Control For Time-varying Situations

Seoung Bum Kim, Korea University,

sbkim1@korea.ac.kr

,

Seulki Lee

In modern manufacturing systems containing the complexity and variability of

processes, appropriate control chart techniques that can efficiently handle the

nonnormal and nonlinear processes are required. In this talk, I will present some

recently developed multivariate control charts to handle both nonnormal and

time-varying process situations.

3 - On The Use Of Support Vectors For Time Series

Pattern Discovery

Mustafa Gokce Baydogan, Bogazici University, Istanbul, Turkey,

mustafa.baydogan@boun.edu.tr,

Mehmet R Kamber,

Erhun Kundakcioglu

Similarity search and classification on time series (TS) databases has received

great interest over the past decade. The definition of similarity between TS is a

major problem in this context. Nearest-neighbor (NN) classifers are widely used

for TS classification but these approaches compute the similarity over the whole

TS which might be problematic with the long TS and relatively short features of

interest. Moreover, these classifiers are not directly interpretable as they do not

describe why a TS is assigned to a certain class. This study utilizes margin

maximization to discover the regions of the time series that have potentially

representative patterns related to the classification task.

4 - Machine Learning For Predicting Heart Failure Readmission

Wei Jiang, Research Assistant, Johns Hopkins University,

3400 N Charles St, Baltimore, MD, 21218, United States,

wjiang1990@gmail.com,

Scott R Levin, Lili Barouch,

Frederick Korley, Sauleh Ahmad Siddiqui, Diego A. Martinez,

Matthew Toerper, Sean Barnes, Eric Hamrock

Predicting risk of heart failure (HF) readmission has gained increasing attention,

with existing studies mainly using administrative data. We will focus on using

clinical data from EMR for predicting HF readmission by doing pattern

recognition with time series clinical data. We will then use classification models

for predicting the drivers of readmission.

T

E C H N I C A L

S

E S S I O N S

Rooms and Locations /Tracks

All tracks / technical sessions will be held in the Music

City Center and Omni Hotel. Room numbers are shown

on the Quick Reference and in the Technical session

listing.