Overcoming the Challenges in Mobile Crowd

SensingNishadh AluthgeDepartment of Computer ScienceUniversity of Helsinki

Mobile Data Analytics seminar, 2016 Spring

Outline

• Introduction to Mobile Crowd Sensing (MCS)• Mobile Crowd sensing Process• Challenges in Mobile Crowd Sensing• Solutions for the Challenges• Conclusion• References

Sensing with Mobile devices

• Large population: 1.86 billion smart phone users.

• Sensors embedded in mobile phones: – GPS, accelerometer, gyroscope, ambient

light, proximity, microphone, and camera sensors

• Mobile sensing devices are pervasively available and are a rich and inexpensive source of sensor data

Mobile Crowd Sensing

Mobile crowd sensing (MCS) refers to applications that leverage consumer mobile devices (GPS, smart phones, and car sensors) to collect and share data about the user or the physical world, either interactively or autonomously, towards a common goal

Mobile Crowd Sensing [Cont’d]

Participatory SensingOpportunistic

Sensing

Users manually determine how, when, what, where to sample.

Takes random sample which is application defined.

Easy to gather large amount data in small time.

Filtering Data by Handling Privacy Issues & Localization.

Dataset is ready for research/ take actions

Users actively engage in the data collection activity.

Mobile Crowd Sensing Process

Challenges in MCS• Privacy ThreatsArises due to sharing of personal information/ habits– Active privacy breach• Narrow tasking: Isolating the participants with

conditions• Selective tasking: Selective task assignment • Local eavesdropping: Intercepting wireless

communication

– Non-active privacy breach• Task tracing attacks: By linking tasks • location-based inference attacks: Frequent spatial

tasks

Challenges in MCS [Cont’d]

• Data reliability issues– False data by users– Critical for the results

• Heterogeneity of mobile platforms– Different mobile platforms (Android, IOS etc.)– Need for separate application development

• Need for separate application installation–Multiple applications need to be installed– Reduces performance

Challenges in MCS [Cont’d]

• High network bandwidth utilization– Core network gets congested– Ex: Video sharing

• Resource limitations on mobile environment– Limited computation, signal processing

power– Higher power usage

Solutions for MCS challenges

• Actions against privacy threats– Processing data locally (Soundsense [1])– Anonymity• PseudonymsUses set of aliases + authentication in communication• Connection anonymizationThe onion routing [2]Mix master [3]

• Actions against privacy threats– Policy based privacy preferences• Limited/ selective participation [6]• Cost (incentive) based task assignment [7]

Solutions for MCS challenges

• Improving Location Reliability (ILR) [8]

Solutions for MCS challenges

– Validates data with minimum effort

– GPS and Bluetooth scans

– Photo tasks & sensing tasks

– Photo selection phase• Graph generation• A set of photo tasks

will be selected (node degree > x)

• Validate photo tasks

– Transitive trust phase: • Extends trust to

collocated data points in Bluetooth scans

Photo tasks

Sensing tasks

• VM based crowd sensing architecture

Solutions for MCS challenges

• VM based crowd sensing architecture [9]– A single application in mobile side– Processing in cloudlets– MAVM aggregates data for optimization– A single copy of data is available for multiple

MCS applications– Solves heterogeneity of platforms, need to

install separate applications, overcoming the resource limitations in mobile devices

– Proxy VMs reduces network bandwidth and network latency

Solutions for MCS challenges

• Mobile crowd sensing is a better way to gather data

• Privacy Threats, Data reliability issues, Heterogeneity of mobile platforms, Need for separate application installation, High network bandwidth utilization, Resource limitations on mobile environment

• Anonymity, Policy based task assignment, VM based MCS architecture, Improving Location reliability

• The solutions depends on the MCS application.

Conclusion

References[1] Lu, H., Pan, W., Lane, N.D., Choudhury, T. and Campbell, A.T., 2009, June. SoundSense: scalable sound sensing for people-centric applications on mobile phones. [2] Dingledine, R., Mathewson, N. and Syverson, P., 2004. Tor: The second generation onion router. Naval Research Lab Washington DC.[3] Mller, U., Cottrell, L., Palfrader, P. and Sassaman, L., 2003. Mixmaster Protocol Version 2. Draft. July, available at: www. abditum.com/mixmaster-spec. txt.[4] Christin, D., Reinhardt, A., Kanhere, S.S. and Hollick, M., 2011. A survey on privacy in mobile participatory sensing applications. Journal of Systems and Software, 84(11), pp.1928-1946[5] Ghinita, G., 2013. Privacy for location-based services. Synthesis Lectures on Information Security, Privacy, & Trust, 4(1), pp.1-85.[6] Pournajaf, L., Xiong, L., Garcia-Ulloa, D.A. and Sunderam, V., 2014. A survey on privacy in mobile crowd sensing task management. Technical Report TR-2014-002, Department of Mathematics and Computer Science, Emory University.

References[7] Riahi, M., Papaioannou, T.G., Trummer, I. and Aberer, K., 2013, March. Utility-driven data acquisition in participatory sensing. In Proceedings of the 16th International Conference on Extending Database Technology (pp. 251-262). ACM.[8] Talasila, M., Curtmola, R. and Borcea, C., 2013, April. Improving location reliability in crowd sensed data with minimal efforts. In Wireless and Mobile Networking Conference (WMNC), 2013 6th Joint IFIP (pp.1-8). IEEE[9] Xiao, Y., Simoens, P., Pillai, P., Ha, K. and Satyanarayanan, M., 2013, February. Lowering the barriers to large-scale mobile crowdsensing. In Proceedings of the 14th Workshop on Mobile Computing Systems and Applications (p. 9). ACM.

Any Questions ?

Thank you !!!

Evaluation - ILR

• Total photo tasks were 1784 and 204 with Bluetooth scans.

• All the photo tasks with Bluetooth scans were considered to be manually validated.

• 45 photos were detected to be fake.• Among them 16 photo tasks were having

Bluetooth scans with false location claims which were provided by 10 malicious users.

• These 16 photo tasks were fed in to ILR to check ILR performance.

Evaluation - ILR

In order to detect 40% of malicious users ILR only required validating 11% of total photos (204 out of 1784)