Ella Peltonen is defending her PhD thesis on crowdsensed mobile data analytics under the supervision of Professor Sasu Tarkoma at the University of Helsinki. The document discusses analyzing data collected from over 850,000 mobile devices to gain insights into smartphone usage and energy optimization. Key findings include that Wi-Fi signal strength, temperature, and screen brightness significantly impact battery life. The thesis also examines how mobile usage varies by geographic region and demographic factors. Recommendations are provided to help users improve battery life and assist developers in building more energy-efficient applications.
2. www.helsinki.fi/yliopisto
Lectio Praecursoria
• Smartphones as a rich data source: terminology
and motivation
• Analytics for crowdsensed data: main findings
and contributions of this thesis
• Impact of the mobile research: practical
applications and future work
3. www.helsinki.fi/yliopisto
Mobile Devices
• Mobile devices: everyday companion devices
• Multifunctional smartphones: camera, game consol,
maps, navigation, communications, etc.
• Carried along almost everywhere, largely personal
• Efficient in computational power, storage space, and
networking capabilities
12. www.helsinki.fi/yliopisto
Mobile Crowdsensing
• Automatic data collection process, no need for user
interaction (crowdsourcing)
• Easy access to large amounts of data in the wild
• Necessity to clean misreadings and default values,
and understand machine-produced data
• Need for Big Data processing tools, e.g., distributed
computing platforms and effective algorithms
13. www.helsinki.fi/yliopisto
Mobile Crowdsensing
• System optimization
• Recommendations for energy efficiency
• Combinations of system settings and
subsystem variables
• Long-term system usage
• Application trends for better application
recommendations
• Influence of demographics, geographics, and
culture to mobile usage
14. www.helsinki.fi/yliopisto
Carat Mobile Data Collection
Since 2012 collected mobile data from over 850,000
unique devices:
● Android and iOS, available in app stores
● Worldwide, circa 200 countries
● 250 million measurements (and more is coming)
● 2 TB in binary objects
● Runnable by Apache Spark in e.g. 10 x ~60GB
RAM, 8 cores Amazon EC2 VMs
● Coarse-grained measurements: based on 1%
battery change
15. www.helsinki.fi/yliopisto
Smartphone Energy Optimization
• Complex interdependencies between system
settings and subsystem variables
• System state: understand setting combinations as
a whole instead of single attributes
• Energy models present energy impact and
transitions between system states
16. www.helsinki.fi/yliopisto
Energy Impact of Individual
Attributes
• CPU has the strongest
impact
• Followed by motion /
stationary
• Screen brightness
contribution relatively low
Attribute Info gain
17. www.helsinki.fi/yliopisto
Energy Impact of Attribute
Combinations
• In combinations, some
attributes of lower impact
alone gain more effect,
e.g.
• Screen brightness
• Battery voltage
• Temperature
• Wi-Fi attributes
Attributes Info gain
18. www.helsinki.fi/yliopisto
Selected Findings of Energy
Consumption Analysis
• Wi-Fi signal strength dropping one bar can
reduce battery life over 13%
• High temperature can cause
even 50% battery loss
• High temperature is not always
related to high CPU load
• Automatic screen brightness is, in the most
cases, better than manual setting
• In addition to CPU, battery temperature and
distance traveled together offer a good predictor
for battery lifetime
27. www.helsinki.fi/yliopisto
Energy Recommendations
• Aims to provide actionable
feedback for smartphone users
• How to adjust user-
changeable system settings
• Helps developers understand
how their applications behave in
the wild
• Problematic subsystem
variables and conditions
28. www.helsinki.fi/yliopisto
Mobile Usage Context
• Device context: technical limitations, system
optimization, network technologies available etc.
• User context: demographic factors reflect needs for
smartphone functionality
• Crowdsensing learns
about the device;
user requirements
surveyed with
questionnaires
32. www.helsinki.fi/yliopisto
Developer Perspective
• Avoiding early stage energy bugs, guaranteeing
compatibility for different platforms, network
infrastructures, etc.
• Marketing and targeting: smartphone usage varies
through different countries and socio-economic
backgrounds
33. www.helsinki.fi/yliopisto
Research Impacts
• Crowdsensed data can be used instead of
laboratory measurements when considering real-
life mobile usage
• Provides a possibility for independent, large-scale
studies not related to marketing companies or
manufacturers
• Energy perspective and different usage habits
around the world have clear design implications
for application development
34. www.helsinki.fi/yliopisto
Summary of Contributions
• Crowdsensed data collection provides a tool to
describe smart device usage in the wild
• Statistical analysis and large-scale machine
learning provide new insights to smartphone usage
• Learning results can thus be translated into
actionable recommendations to gain the most
benefit for mobile users and developers
36. www.helsinki.fi/yliopisto
Publications
Energy Modeling of System Settings: A
Crowdsourced Approach, Percom 2015
Constella: Recommending System Settings the
Crowdsourced Way, Pervasive and Mobile
Computing Feb 2016
Carat: Collaborative Energy Diagnosis for Mobile
Devices, SenSys 2013
How Carat Affects User Behavior: Implications for
Mobile Battery Awareness Applications, CHI 2014
The Company You Keep: Mobile Malware Infection
Rates and Inexpensive Risk Indicators, WWW 2014