Mobile Crowdsourcing: From Theory to Practice

Preface
Contents
Part I Introduction
Crowdsourcing as a Future Collaborative Computing Paradigm
1 Definition and History
1.1 HPU and CPU
1.2 Basic Components
1.3 History
2 Crowdsourcing Events in Recent History
2.1 Help Find Jim Gray (2007)
2.2 Malaysia Airlines Flight MH 370 (2014)
2.3 DARPA Network Challenges (2009)
2.4 Tag Challenges (2012)
2.5 Kasparov vs. IBM Deep Blue (1997)
2.6 A Big Picture: Human vs. Machine
3 Crowdsourcing Overview
3.1 Workflow of Crowdsourcing
3.2 Types of Crowdsourcing
4 Platform
4.1 Amazon Mechanical Turk
4.2 Crowd4U
4.3 gMission
4.4 UpWork
4.5 CrowdFlower
5 Sample Applications
5.1 Image Processing
5.2 Commonsense Knowledge
5.3 Smart City
5.4 Other Science Projects
5.5 Mixed HPU and CPU Applications
6 Algorithmically and Theoretically Challenging Issues
6.1 Paradigm
6.1.1 Sequential Implementation
6.1.2 Parallel Implementation
6.1.3 Divide and Conquer
6.2 Multi-Armed Bandit (MAB)
6.3 Incentive Mechanisms
7 Opportunities and Future Directions
7.1 Beyond Simple Workflows
7.2 Beyond Simple Worker Selection
7.3 Beyond Independent Workers
7.4 Beyond Simple Training
7.5 Beyond Simple Interactive Mode
7.6 AI Applications
7.7 Crowdsourcing 2.0
8 Conclusion
References
Urban Mobility-Driven Crowdsensing: Recent Advances in Machine Learning Designs and Ubiquitous Applications
1 Introduction
2 An Overview of Urban Mobility-Driven Crowdsensing
3 Advancing Machine Learning Designs for UMCS
3.1 Machine Learning Advances in Crowdsensed Signal Reconstruction
3.2 Machine Learning Advances in Understanding Crowd Mobility Distributions
4 Expanding Ubiquitous Use Cases for UMCS
4.1 Indoor Crowd Detection and Group Identification
4.2 Urban Mobility Reconfiguration with Crowdsourced Information Fusion
5 Conclusion
References
Part II Key Technical Components: User Recruitment and Incentive Mechanisms
Unknown Worker Recruitment in Mobile Crowdsourcing
1 Introduction
2 Related Work
3 System Model and Workflow
3.1 System Model
3.2 System Workflow
4 Unknown Worker Recruitment Scheme
4.1 Modeling and Formulation
4.2 Algorithm Design
4.3 Theoretical Analysis
4.4 Extansion: Budget-Limited UWR Scheme
5 Privacy-Preserving Unknown Worker Recruitment Scheme
5.1 DP-MAB Model
5.2 Problem Formulation
5.3 The DPF Algorithm
5.4 Performance Analysis of DPF Algorithm
5.5 DPU Algorithm
5.6 Performance Analysis of DPU algorithm
6 Conclusion
References
Quality-Aware Incentive Mechanism for Mobile Crowdsourcing
1 Introduction
2 Related Work
3 Quality-Aware Incentive Mechanisms for MCS Systems
3.1 System Overview
3.2 Auction Model
3.3 Design Objective
3.4 SRC Auction
3.4.1 Mathematical Formulation
3.4.2 Mechanism Design
4 Quality-Aware Incentive Mechanism Considering the Bid Privacy for MCS Systems
4.1 System Overview
4.2 Aggregation Method
4.3 Auction Model
4.4 Design Objective
4.5 Mathematical Formulation
4.6 Mechanism Design
5 Conclusion
References
Incentive Mechanism Design for Mobile Crowdsourcing Without Verification
1 Introduction
1.1 Motivations
1.2 Key Challenges
1.3 Chapter Outline
2 Model
2.1 Workers' Decisions and Payoffs
2.1.1 Task and Workers
2.1.2 Worker Effort Exertion Strategy
2.1.3 Worker Solution Reporting Strategy
2.1.4 Incentive Mechanism
2.1.5 Worker Payoff
2.2 Platform's Decisions and Payoff
2.2.1 Platform Decisions
2.2.2 Platform Payoff
2.3 Platform–Worker Interaction
3 Approaches to Worker Heterogeneity
3.1 Motivating Examples and Key Questions
3.2 Solution: Majority Voting Mechanism
3.3 Results and Insights
4 Approaches to Worker Collusion
4.1 Motivating Examples and Key Questions
4.2 Solution: Truth Detection Mechanism
4.3 Results and Insights
5 Approaches to Information Incompleteness
5.1 Motivating Examples and Key Questions
5.2 Solution: Randomized Learning Mechanism
5.3 Results and Insights
6 Approaches to Information Asymmetry
6.1 Motivating Examples and Key Questions
6.2 Solution: Bayesian Persuasion Mechanism
6.3 Results and Insights
7 Conclusion and Open Problem
7.1 Future Challenges and Open Issues
7.1.1 Joint Optimization of Information Elicitation and Aggregation
7.1.2 Competitive Market
7.1.3 Worker Bounded Rationality
7.1.4 Worker Privacy and Moral Issues
7.2 Conclusion
References
Part III Key Technical Components: Task Allocation
Stable Worker–Task Assignment in Mobile Crowdsensing Applications
1 Introduction
2 Background
2.1 Worker–Task Assignment in Mobile Crowdsensing
2.2 Matching Under Preferences
3 Why Should We Care About Stability in MCS?
4 Stable Task Assignments in Different MCS Applications
4.1 Participatory MCS
4.2 Opportunistic MCS
4.3 Hybrid MCS
5 Conclusion and Open Problems
References
Spatiotemporal Task Allocation in Mobile Crowdsensing
1 Introduction
2 Optimized Allocation of Time-Dependent Tasks for Mobile Crowdsensing
2.1 Problem Statement
2.2 System Overview
2.3 Problem Formulation
2.4 Task Allocation Algorithm
2.5 Performance Evaluation
3 Heterogeneous User Recruitment of Multiple Spatiotemporal Tasks
3.1 Problem Statement
3.2 System Overview
3.3 Problem Formulation
3.4 Model Analysis
3.4.1 Heterogeneous Task Priority Model
3.4.2 Platform Payment Incentive Model
3.4.3 User-Contributed Task Coverage Ratio Model
3.4.4 Binary-Based Representation of Level
3.5 HURoT Problem-Solving Approaches
3.5.1 Utility Function with Dual Objectives
3.5.2 Utility-Based User Recruitment (UURe)
3.5.3 Level-First and Utility-Based User Recruitment (L-UURe)
3.5.4 Global Level-First and Utility-Based User Recruitment (GL-UURe)
3.6 Performance Evaluation
3.6.1 Experiment Settings
3.6.2 Experimental Results and Analysis
4 Conclusion
References
Part IV Key Technical Components: Data Inference
Joint Data Collection and Truth Inference in Spatial Crowdsourcing
1 Introduction
1.1 Challenges and Motivations
2 Model of Truth Inference and Task Allocation
2.1 System Overview
2.2 Truth Inference
2.2.1 Numerical Task
2.2.2 Categorical Task
2.3 Task Allocation
2.4 Process of Crowdsourcing System
3 Online Expertise-Aware Truth Inference
3.1 Maximum Likelihood Numerical Inference
3.2 Expectation Maximization Categorical Inference
3.3 Algorithm Design for Truth Inference
4 Online Location-Aware Task Allocation
4.1 Probability Improvement-Based Allocation
4.1.1 Numerical Tasks
4.1.2 Categorical Tasks
4.2 Entropy-Reduction-Based Allocation
4.2.1 Numerical Tasks
4.2.2 Categorical Tasks
4.3 Algorithm Design for Task Allocation
5 Performance Evaluation
5.1 Dataset and Settings
5.1.1 Dataset
5.1.2 Parameter Settings
5.1.3 Comparison Algorithms
5.1.4 Evaluation Metric
5.2 Results of Truth Inference
5.3 Results of Task Allocation
5.4 Results of Running Time
5.5 Results on Larger Dataset
6 Chapter Summary
References
Cost-Quality Aware Compressive Mobile Crowdsensing
1 Background
2 System Model and Problem Statement
2.1 System Model
2.2 Data Inference
2.3 Importance Assessment
2.4 Cost Assessment
2.5 Quality Assessment
2.6 Problem Formulation
3 Advanced Cell Selection Strategies in CCS
3.1 Randomized Sampling Strategy
3.1.1 Recovery Accuracy Prediction Based on Regularized Column Sum
3.1.2 CACS via Convex Optimization
3.2 Active Sampling Strategy with Multiple Steps
3.2.1 Use Case Study
3.2.2 Cost Estimation
3.2.3 Cost–Quality Beneficial Cell Selection
3.3 Active Sampling Strategy Based on Bipartite Graph
3.3.1 Representing Matrix Factorization Based on Bipartite Graph
3.3.2 Sampling to Form a Complete and Robust Linear System
4 Evaluation
4.1 Experimental Setup
4.1.1 Datasets
4.1.2 Baselines
4.2 Experimental Results
4.2.1 Errors of Inferred Value
4.2.2 The Number and Total Costs of Selected Cells
5 Summary
References
Part V Key Technical Components: Security and Privacy
Information Integrity in Participatory Crowd-Sensing via Robust Trust Models
1 Introduction
2 Architecture for Participatory MCS
3 Security Threats and Challenges
3.1 Types of Dishonest Behaviors
3.2 Cold Start Problem and Other Challenges
3.3 Categories of Vulnerabilities and Attack Types:
4 Quality and Quantity Unified Architecture for Secure and Trustworthy Crowd-sensing
4.1 Robust Quality of Information Model
4.1.1 Posterior Estimation of Probability Masses
4.1.2 Non-linear Weighing of Probability Masses:
4.1.3 Link Function
4.2 Robust User Reputation Scoring Module
4.2.1 Modified Link Functions
4.2.2 One-Hot Encoded Sum
4.2.3 Output Activation and Classification Criterion
5 Analytical Case Study
6 Conclusion
References
AI-Driven Attack Modeling and Defense Strategies in Mobile Crowdsensing: A Special Case Study on Fake Tasks
1 Introduction
2 Background on Mobile Crowdsensing
2.1 Use Cases of MCS
2.2 System Architecture of MCS
2.3 Quality of Service in MCS
3 Security and Threat Models in MCS
3.1 Threat Models in MCS
4 AI-Driven Attack Anticipation in MCS
4.1 Fake Task Injection Modeling
4.2 Types of Task Movement
4.2.1 Zone-Free Task Movement (ZFM)
4.2.2 Zone-Limited Task Movement (ZLM)
4.3 Self-organizing Feature Map Implementation for Attack Modeling
4.4 Region-Based SOFM Structure
4.5 Locally Reconfigurable SOFM for More Impactful Attack Region Selection
5 AI-Driven Defense Strategies in MCS
5.1 AI-Backed Legitimacy Detection
5.2 Machine Learning Model Development to Increase the Performance of Legitimacy Detection
6 Conclusion
References
Traceable and Secure Data Sharing in Mobile Crowdsensing
1 Introduction
2 Related Work
2.1 Mobile Crowdsensing
2.2 Privacy-Enhancing Techniques for Mobile Crowdsensing
3 Traceable and Privacy-Preserving Non-interactive Data Sharing (TIDS) Scheme
3.1 Problem Statement
3.1.1 System Model
3.1.2 Threat Model
3.1.3 Design Goals
3.2 Preliminaries
3.2.1 Bilinear Pairings
3.2.2 Access Structure
3.3 The TIDS Scheme
3.3.1 The TIDS Framework
3.3.2 The Detailed Description of TIDS
3.4 Security Analysis
3.5 Performance Analysis
3.5.1 Theoretical Analysis
3.5.2 Experimental Evaluations
4 Conclusion and Future Work
References
User Privacy Protection in MCS: Threats, Solutions, and Open Issues
1 Introduction
2 User Privacy Threats and Requirements
2.1 Threat Model
2.2 Privacy Attacks
2.3 Privacy Threats
2.3.1 Privacy Threats from Task
2.3.2 Privacy Threats from Data
2.4 Privacy Leakage in the Whole Data Flow Process
2.4.1 Privacy Leakage in Task Allocation
2.4.2 Privacy Leakage in User Incentive
2.4.3 Privacy Leakage in Data Collection
2.4.4 Privacy Leakage in Data Processing and Publishing
2.5 Requirements for User Privacy
3 Privacy Protection Technologies
3.1 Anonymization-Based Technologies
3.1.1 Generalization
3.1.2 Suppression
3.2 Perturbation-Based Technologies
3.2.1 Randomized Response
3.2.2 Differential Privacy
3.3 Encryption-Based Technologies
3.3.1 Fully Homomorphic Encryption
3.3.2 Partially Homomorphic Encryption
4 Privacy Protection for Mobile Users
4.1 User Privacy Protection in Task Allocation
4.2 User Privacy Protection in Incentive
4.3 User Privacy Protection in Data Collection and Publishing
5 Open Issues for Mobile User Privacy Protection
5.1 Full Lifecycle Privacy Protection Framework
5.2 Breaking the Privacy-Overhead-Utility Trilemma
5.3 Incorporation of Novel Privacy-Preserving Computing Technologies
5.4 Privacy Protection for Users' Fresh Time-Series Data
6 Conclusion
References
Part VI Applications
Crowdsourcing Through TinyML as a Way to Engage End-Users in IoT Solutions
1 Introduction
2 Strategies for End-User Engagement
3 Basic Concepts of TinyML
3.1 Building a TinyML Application
3.2 TensorFlow Lite Micro
3.3 The Edge Impulse Platform
4 Example Applications with TinyML
5 TinyML on Device Development
6 Beehive-Application
7 Conclusions
References
Health Crowd Sensing and Computing: From Crowdsourced Digital Health Footprints to Population Health Intelligence
1 Introduction
2 Preliminaries and Fundamentals
2.1 Fundamentals for Population Health and Epidemiology
2.2 Crowd Sensing and Its Applications in Health Care
2.3 AI Technologies as Enablers of HCSC
3 Conceptual Framework for HCSC
4 Case Study: Compressive Population Health
4.1 Approach
4.2 Experimental Results
5 Research Opportunity and Proposal for Future HCSC
6 Non-scientific Considerations for HCSC: Privacy, Ethics, and Security
7 Conclusion
References
Crowdsourcing Applications and Techniques in Computer Vision
1 Introduction
2 Computer Vision
2.1 Machine Learning Applied to Computer Vision
3 Crowdsourcing Applications in Computer Vision
3.1 Computer Vision Datasets
3.2 Labelling Software
3.3 Crowdsourcing Use Cases
4 Crowdsourcing Data Aggregation and Evaluation
5 Concluding Remarks
References
Mobile Crowdsourcing Task Offloading on Social Collaboration Networks: An Empirical Study
1 Introduction
2 Related Work
3 Preliminary Understanding by Questionnaire
4 Further Experiment: Field Experiments and Analysis
4.1 Experimental Settings
4.2 Study I: The Basic Situation
4.2.1 Basic Observation in Task Offloading
4.2.2 The Impact of Punitive Measures
4.3 Study II: Task Offloading Pattern Investigation
4.3.1 The Patterns of MCS Task Offloading
4.3.2 Multi-hop MCS Task Offloading
4.4 Study III: Incentive Mechanism in Task Offloading
4.4.1 Effect of Bid Incentive Rewards
4.4.2 Incentive Reward-Sharing Mechanism
5 Conclusion
References