02

8 downloads 0 Views 3MB Size Report
Jiangtao Wang et al. Energy Saving Techniques in Mobile Crowd Sensing: Current State and. Future Opportunities. IEEE Communications Magazine. (in press ...
Jiangtao Wang, Peking University

2018/02 @UNSW, Sydney

Outline

Backgrounds

1 2

New Perspectives

3

Future Work

Power of the Crowds

Lior Zoref TED Talk

Crowds + X  New Paradigm or Research Topic

Urban Sensing MCS

Creative Work

Software Engineering

Crowdsourcing

Knowledge Graph Refinement

Data Mining

Mobile Crowd Sensing (MCS): Research Background Human mobility, embedded sensor Collect and report data in mobile device

Cloud Server

MCS 任务 任务 任务 Task

Data integration Air quality Noise level

Flow of citizens

Traffic congestion status

Jiangtao Wang et al. Energy Saving Techniques in Mobile Crowd Sensing: Current State and Future Opportunities. IEEE Communications Magazine. (in press, IF:10.4).

Framework of MCS Research Research on MCS Applications Environment Sensing

Social Sensing

Infrastructure Sensing

Research on Core Supporting Techniques Task creation

Task allocation

Task execution

Data integration

End-user programming

Quality control

Energy saving

Missing data inference

Cost control

Privacy preserving

Micro-Task design

visualization

My Early Research Experience in MCS (PhD student)

MCS application research: MCS-based queue time estimation system • Queue behavior detection with accelerometer sensor. • Using acoustic context to improve accuracy  Task creation techniques: From the perspective of software engineering • Help task creators with or without professional programming skills to create their tasks efficiently. • • •

J Wang, Y Wang, D Zhang, L Wang, C Chen, J Lee, Y He: Real-Time and Generic Queue Time Estimation based on Mobile Crowdsensing. Frontiers of Computer Science 12/2016. J Wang, Y Wang, S Helal, D Zhang: A Context-Driven Worker Selection Framework for Crowd-Sensing. International Journal of Distributed Sensor Networks 03/2016; 2016(3):1-16. J Wang, Y Wang, L Wang, Y He (2017). GP-selector: a generic participant selection framework for mobile crowdsourcing systems. World Wide Web, 1-24.

Framework of MCS Research Research on MCS Applications Environment Sensing

Social Sensing

Infrastructure Sensing

Research on Core Supporting Techniques Task creation

Task allocation

Task execution

Data integration

End-user programming

Quality control

Energy saving

Missing data inference

Micro-Task design

Cost control

Privacy preserving

visualization

The focus of the recent work (2016~now)

Task Allocation in MCS: An Overview Various data quality measures, incentive models, optimization goals and constraints… Single-objective vs Multi-objective

Goal & Constraint

Problem formulation of MCS task allocation Online vs Offline

Single task vs Multi-task

Hot topics in top venues of recent years: CSCW, UBICOMP, TMC, ICDE, INFOCOM….

Outline

Backgrounds

1 2

New Perspectives

3

Future Work

Our Recent Work: New Perspective in MCS Task Allocation

P1 P2 P3

• From single task to multi-task • Multi-task allocation with single-task quality considered

• Hybrid task allocation

Research Perspective 1:From single-task to multi-task  “single task assumption “ in existing work:tasks are independent  popularity of MCS  resource competition (e.g., budget and participants sensing bandwidth) among multiple tasks,“single task assumption "does not hold any longer

Multi-task allocation in MCS

Scenario A One Organizer Multiple Tasks

Goal:overall utility Constraint: total budget Naï ve method based on divided budget: overall utility is not optimal

Scenario B Multiple Organizers Multiple Tasks

Goal:overall utility Constraints: sensing bandwidth + task budget Single task oriented approach  participant overload

Research Perspective 1:From single-task to multi-task Basic idea: (1) participant mobility prediction; (2)overall utility modeling;(3) greedy selection with submodular theory (near-optimal)

 Jiangtao Wang, Yasha Wang, Daqing Zhang, Haoyi Xiong, Leye Wang, Helal Sumi, Yuanduo He, Feng Wang: Fine-Grained Multi-Task Allocation for Participatory Sensing with a Shared Budget. IEEE IOT JOURNAL, VOL. 3, NO. 6, DECEMBER 2016 (IF: 7.6)  Jiangtao Wang, Yasha Wang, Daqing Zhang, Feng Wang, Yuanduo He, Liantao Ma: PSAllocator: MultiTask Allocation for Participatory Sensing with Sensing Capability Constraints. In Proceedings of the 2017 ACM Conference on Computer Supported Cooperative Work and Social Computing (CSCW 2017). ACM. (CCF A)

Our Recent Work: New Perspective in MCS Task Allocation

P1 P2 P3

• From single task to multi-task • Multi-task allocation with single-task quality considered

• Hybrid task allocation

 Research Perspective 2:multi-task allocation with single-task quality considered  Existing work optimize overall utility without considering single task quality  Low quality of single task  MCS service unavailable  resource wasting. (e.g., navigation service: 20% coverage of traffic status monitoring) Optimizing overall utility with single-task quality considered (set a minimum quality threshold and re-define overall utility) Baseline 1

Baseline 2

when utility increase of each item is zero random selection

using relaxed utility

Shortcoming: Overall utility may not be near-optimal

Shortcoming: Resource wasting (the quality of tasks with assigned participant does not reach to the minimum threshold

Basic idea:assign resources to the tasks that can finally reach to the minimum quality requirement (avoid resource wasting)

Research Perspective 2:multi-task allocation with single-task quality considered

Basic idea: participant mobility prediction+ descent greedy approach •

• •

Assign all resources by “ignoring” the bandwidth constraint of each participant. Delete task-participant pair one by one with minimum overall utility reduction. For tasks failing to reach minimum quality: abandon task and release corresponding resources

Jiangtao Wang, Yasha Wang, Daqing Zhang, Feng Wang, Haoyi Xiong, Chao Chen, Qin Lv, Zhaopeng Qiu. Multi-Task Allocation in Mobile Crowd Sensing with Individual Task Quality Assurance. IEEE Transactions on Mobile Computing, 2018.

Our Recent Work: New Perspective in MCS Task Allocation

P1 P2 P3

• From single task to multi-task • Jointly consider individual task and overall multi-task utility

• Hybrid Task Allocation

Research Perspective 3: Hybrid Task Allocation (1/3)

Figures borrowed from Guo et al

Opportunistic Mode

+ low cost, less intrusive

- tasks may not be completed

Hybrid mode

Participatory Mode

+ completion assured - intrusive, high cost

Research Perspective 3: Hybrid Task Allocation (2/3)

Payment: • Opportunistic worker: fixed reward • Participatory worker: in proportion to movement distance

 Constraint: total budget  Goal: maximize the number of completed tasks

Naïve method: divide the budget into two parts  may not find a optimal division Hence, we need to design more sophisticated method----jointly

consider and optimize task allocation of these two modes.

Research Perspective 3: Hybrid Task Allocation (3/3) Key challenge: how to select opportunistic workers by jointly considering future task allocation for participatory workers ?

Prefer to select OPP workers Visit more task locations

Visit task locations where the participatory workers are sparsely distributed

Submitted to IEEE TMC (under review)

Outline

Backgrounds

1 2

New Perspectives

3

Future Work

New Trend in Crowd Computing ?

Crowd Learning 4 Crowd

Crowd 4 Learning

Learning Jiangtao Wang et al. , Crowd-Assisted Machine Learning: Current Issues and Future Directions, submitted to IEEE Computer (under review).

References 1.

2.

3.

4.

5. 6.

Jiangtao Wang, Yasha Wang, Daqing Zhang, Feng Wang, Yuanduo He, Liantao Ma: PSAllocator: Multi-Task Allocation for Participatory Sensing with Sensing Capability Constraints. In Proceedings of the 2017 ACM Conference on Computer Supported Cooperative Work and Social Computing (pp. 1139-1151). ACM. (CCF A) Jiangtao Wang, Yasha Wang, Daqing Zhang, Feng Wang, Haoyi Xiong, Chao Chen, Qin Lv, Zhaopeng Qiu. Multi-Task Allocation in Mobile Crowd Sensing with Individual Task Quality Assurance. IEEE Transactions on Mobile Computing, 2018. (CCF A) Jiangtao Wang, Yasha Wang, Daqing Zhang, Haoyi Xiong, Leye Wang, Helal Sumi, Yuanduo He, Feng Wang: Fine-Grained Multi-Task Allocation for Participatory Sensing with a Shared Budget. IEEE Internet of things journal, vol. 3, no. 6, 2016. (IF:7.6) Jiangtao Wang, Yasha Wang, Daqing Zhang, Sumi Helal. Energy Saving Techniques in Mobile Crowd Sensing: Current State and Future Opportunities. IEEE Communications Magazine ( IF:10.4). Jiangtao Wang, Yasha Wang, Yafei Wang (2017). CAPFF: A context-aware assistant for paper form filling. IEEE Transactions on Human-Machine Systems, 47(6), 903-908. Jiangtao Wang, Yasha Wang, Leye Wang, Yuanduo, He. GP-Selector: A Generic Participant Selection Framework for Mobile Crowdsourcing Systems. World Wide Web Journal , Springer ,2017. Full text downloading:https://www.researchgate.net/profile/Jiangtao_Wang4

Welcome to visit Peking University