emotionsense: a mobile phones based adaptive platform for
TRANSCRIPT
Kiran K. Rachuri† Mirco Musolesiψ Cecilia
Mascolo†
† Computer Laboratory, University of CambridgeψSchool of Computer Science, University of St. Andrews
Energy-Accuracy Trade-offs in Querying Sensor Data for
Continuous Sensing Mobile Systems
1
Sense continuously
Plays a central role in many context-aware applications
High level classifiers
Energy-Accuracy Trade-offs
Continuous Sensing Mobile
Systems
2
Mobile Phone Limitations
Energy, processing, and
memory constraints
Google Nexus One, and
HDC HD2 are equipped
with1GHz processor and
512MB RAM
Energy is still a scarce
resource
3
Sensor Sampling
4
Time
Sleep Sense Sleep Sense Sleep
Event
s
Time
Sleep Sense
Event
s
Sense SleepSleep
Sensor Sampling Issues
Continuous sampling degrades battery life
Long sleep durations result in loss of sensor data
Different sensors have different requirements
Accuracy varies with sensors and classifiers
5
Sampling Interval
6
Sleep Sense Once
Sleep 0 ∞Minimum
Sampling
Interval
Maximum
Sampling
Interval
Generally
constant
for a
sensor
Design Methodology
7
Missable Event
Sleep
Not all events are important
E.g.: Microphone recording
when there is no audible sound
•Classify events as Unmissable
and Missable
•Use functions to control the sleep
interval
Back-off Function
E.g.: f(x) = 2x, where x is sleep interval
UnMissable Event
Advance Function
E.g.: f(x) = x/2, where x is sleep
interval
Sense
Back-off and Advance Functions
Type Back-off function Advance function
Linear
Quadratic
Exponential
Minimum N/A Minimum interval
Maximum Maximum
interval
N/A
k
x
2x
xe
x
8 x: sleep
interval
Dynamic Adaptation
9
Dynamically switch functions from least to most
aggressive
Missable Event Sleep
Sequence Count
Sense
Linear back-off function
Quadratic back-off
function
Exponential back-off
function
Update
Sleep
Interval
< Linear
Threshold
< Quadratic
Threshold
> Quadratic
Threshold
Evaluation – Sensor Traces
Ground truth traces - 10 users for 24 hours
Continuous sampling of Accelerometer, Bluetooth, and Microphone sensors
Events in the trace files are classified as “missable” and “unmissable” events
The classifiers are based on the EmotionSensesystem10
Event Classification
Microphone
Missable: Silence data
Unmissable: Some audible voice data
Based on silence detection technique in EmotionSense
Bluetooth
Missable: No change in co-location
Unmissable: Change in co-location
Accelerometer
Missable: Stationary
Unmissable: Moving
11
Results
Bluetooth linear threshold – Dynamic adaptation
algorithm
12
Results
Bluetooth – minimum interval variation
13
Dynamic algorithm is more accurate than
exponential_linear by a factor of 5, whereas the gain ratio
w.r.t. to Energy is only 1.5
Results
Accelerometer – minimum interval variation
14
Difference in accuracy is non-negligible whereas
difference in energy consumption is insignificant
Results
Microphone – minimum interval variation
15
linear_exponential is a better option for microphone
sensor as the energy savings are higher than the benefit
w.r.t. accuracy
Summary
Continuous sensing mobile systems
Function based sensor sampling rate control
Dynamic adaptation
Different sensors have different requirements
16
Discussion
Dynamic adaptation to unknown classifiers
Unknown sensors
Mobile phones differ in capabilities and
resources
17
18
Questions?
Thank You
Kiran Rachuri
EmotionSense
Demo: Mon 27 Sep, Blixen room, 12:30 to 14:00
Talk: Wed 29 Sep 13:30
http://www.cl.cam.ac.uk/research/srg/netos/emotionsense/