Gary M. Weiss and Jeffrey Lockhart Fordham University, New York, NY

voltaireblingData Management

Nov 20, 2013 (3 years and 4 months ago)

72 views

Gary M. Weiss and Jeffrey Lockhart

Fordham University, New York, NY


1

UbiMI

Workshop @ UBICOMP Sept. 8 2012


Mobile sensors becoming ubiquitous


Especially via smartphones


Various architectures are possible ranging
from “smart client” to “dumb client”


Each architecture has pros and cons


Worthwhile to enumerate and compare
alternative architectures

2

UbiMI

Workshop @ UBICOMP Sept. 8 2012

1.
Sensor Collection

2.
Data Processing and Transformation

3.
Decision Analysis/Model Application

4.
Data and Knowledge Reporting


Learning/model generation

Only step 1 is required


3

UbiMI

Workshop @ UBICOMP Sept. 8 2012


Main focus of WISDM lab


Monitors smartphone accelerometer and
uses the data to perform activity recognition


Activities: walk, jog, stairs, sit, stand, lie down


Results available via the Web

4

UbiMI

Workshop @ UBICOMP Sept. 8 2012


Sensor Collection:


Actitracker
client

collects raw accelerometer data for
3 axes 20 times per second and transmits to server


Data Processing and Transformation


Every 10 sec.
server

aggregates raw samples into a
single example described by several dozen features


Decision Analysis/Model Application


Server

applies predictive model to examples; activity
classified and saved to database


Data and Knowledge Reporting


User queries
server

DB any time via web interface


5

UbiMI

Workshop @ UBICOMP Sept. 8 2012

Client Configurations

Responsibility

CC
-
1

Dumb

CC
-
2

CC
-
3

CC
-
4

Smart

1

Sensor Collection









2

Data

Transformation







3

Model Application





4

Reporting



Model Generation

?

?

6

UbiMI

Workshop @ UBICOMP Sept. 8 2012


Mobile devices have CPU power to build models


Only makes sense to build a model on the client device if
will apply it on the client


Thus model construction on device only for CC
-
3 or CC
-
4


In CC
-
1 and CC
-
2 either model hardcoded into client or
downloaded from server


Data mining not always required


Can be done dynamically (on client or server) or statically


Our research shows dynamically generated personal
models outperform general (impersonal) models
1

7

UbiMI

Workshop @ UBICOMP Sept. 8 2012

1

Gary M. Weiss

and Jeffrey W. Lockhart. The Impact of Personalization on Smartphone
-
Based Activity Recognition,
Papers from the AAAI
-
12 Workshop on Activity Context Representation: Techniques and Languages
, AAAI Technical
Report WS
-
12
-
05, Toronto, Canada, 98
-
104.


Resource usage


battery, CPU, memory, transmission bandwidth


Scalability


Support for many mobile devices


Access to data


Researchers and others may want raw data


Transformed data loses information


With raw data can alter features for data mining and
regenerate results



UbiMI

Workshop @ UBICOMP Sept. 8 2012

8


Privacy/Security


Users will want to keep data secure and/or private


User Interface


Users want aesthetics (screen size) & accessibility


Crowdsourcing


Some applications will require a central server in
order to aggregate data from multiple users/devices


Navigation software that tracks traffic

UbiMI

Workshop @ UBICOMP Sept. 8 2012

9


Resource Usage


Unclear
. Resource usage minimized except heaviest
use of transmission bandwidth (power drain)


Scalability


Poor

since maximizes server work


Actitracker’s server can handle 942
simult
. users


Access to Data


Best

since all raw data can be preserved on server


But Actitracker requires 791 MB/month per user.


UbiMI

Workshop @ UBICOMP Sept. 8 2012

10


Privacy/Security:


Poor
: The more data sent the greater the risk


User Interface:


Good
: data and results on server and can be viewed
over Internet


Crowdsourcing


Best
: All data available on server

UbiMI

Workshop @ UBICOMP Sept. 8 2012

11


Similar to CC
-
1 except:


Less data to transmit so bandwidth/energy savings


For Actitracker 95% reduction in data


But more processing which takes up CPU and power


More scalable (less server work)


Less access to data (raw data not available)


Slight improvement in privacy/security (no raw data)


Minimal impact on user interface (results still on server)


Crowdsourcing only on aggregated data




UbiMI

Workshop @ UBICOMP Sept. 8 2012

12


Resource usage:


more processing on the client (more CPU and power);
but only need to transmit results


Much more scalable: server only collects results


Access to data: only results available


Much improved security/privacy


results may not be nearly as sensitive


Can still view results via web
-
based interface


Can only crowdsource on results


UbiMI

Workshop @ UBICOMP Sept. 8 2012

13


About same as CC
-
3


not sending results saves little power


Perfectly scalable: no server


No access to data


Good security/privacy: nothing leaves device


Can only view results on the device


Not accessible from other places and small screens


Cannot even crowdsource results

UbiMI

Workshop @ UBICOMP Sept. 8 2012

14


Resource usage:

unclear


Scalability:


smart client best


Access to data:

dumb client best


Security/Privacy:

smart client best


User Interface:

smart client worst


Centralized Data:

dumb client best


One approach: support multiple architectures



approach taken by our research group

UbiMI

Workshop @ UBICOMP Sept. 8 2012

15


Go to wisdmproject.com


Actitracker should be ready for beta in 1 month


Actitracker.com


Papers available from:


http://www.cis.fordham.edu/wisdm/publications.php


My contact info:


gweiss@cis.fordham.edu

UbiMI

Workshop @ UBICOMP Sept. 8 2012

16