Crowd Sensing: From Pervasive Sensing to Social Behavior Jun Luo - - PowerPoint PPT Presentation

Crowd Sensing: From Pervasive Sensing to Social Behavior

Jun Luo

School of Computer Engineering Nanyang Technological University, Singapore Supported in part by

Complexity Institute

NCUS

Jun Luo (NTU) Crowd Sensing Complexity Community 1 / 17

Outline

1

Introduction

2

Indoor Floor Plan Reconstruction

3

Vehicle-Based Crowd Sensing

4

City-Scale Crowd Sensing

5

The Social Aspects of Crowd Sensing Challenges as A Social Behavior Truthful Scheduling: A Theoretical Problem Scheduling and Incentive Mechanism Design

6

Conclusion

Jun Luo (NTU) Crowd Sensing Complexity Community 2 / 17

Introduction

Crowd Sensing – A Rough Definition Why and what

Smartness of nation/city demands high information availability Existing data gathering infrastructures are stale Relying on the crowd for gathering information is both cost-effective and robust While human can be sensors, they are extended with

• ther sensing devices

Other sensors beside smart phone

Car has numerous sensors SmartWear (e.g., Google Glass) is becoming more and more popular

Jun Luo (NTU) Crowd Sensing Complexity Community 3 / 17

Introduction

Crowd Sensing – What Can We Do With It? It does make us far smarter than we are now Environment monitoring and neighbor surveillance Large scale scene reconstruction Crowdedness and occupancy detection Indoor localization and navigation in large indoor facilities Traffic and parking management Getting better aware of things happening around us

Jun Luo (NTU) Crowd Sensing Complexity Community 4 / 17

Indoor Floor Plan Reconstruction

Infrastructure-Free Indoor Localization Indoor localization is an important aspect

• f smart city, but

Indoor floor plans are often unavailable Some form of deployments, e.g., Wi-Fi APs or BT beacons, are often needed Out solution: geomagnetism assisted crowd sensing Using “twisted” geomagnetic field as location fingerprints Traces sensed by crowd are assembled to form a floor plan

A B C D E

A B C D E A B C D A B C D E 82m 116m

Alice Bob Bob

A B C D E A B C D E Jun Luo (NTU) Crowd Sensing Complexity Community 5 / 17

Indoor Floor Plan Reconstruction

A Concrete Example - Crowd Sampling

(a) First trace (b) Second trace (c) Third trace (d) Fourth trace (e) Fifth trace (f) Sixth trace (g) Seventh trace (h) Eighth trace

Jun Luo (NTU) Crowd Sensing Complexity Community 6 / 17

Indoor Floor Plan Reconstruction

A Concrete Example - Floor Plan Assembling

(a) First trace (b) Second trace (c) Third trace (d) Fourth trace (e) Fifth trace (f) Sixth trace (g) Seventh trace (h) Eighth trace

Jun Luo (NTU) Crowd Sensing Complexity Community 7 / 17

Vehicle-Based Crowd Sensing

Parking Availability Detection by Crowd Do you want to have a map that indicates where the street parking or parking garages are how many places are available for each facility However, the information may not come handy in digital form Street parking may change from time to time Parking garages do not put their available places online Our solution is again based on crowd sensing Crowd can be used to discover changes in parking places Parking garage availability can be inferred by the time spent for parking individual cars

Jun Luo (NTU) Crowd Sensing Complexity Community 8 / 17

City-Scale Crowd Sensing

A Nice Wedding of Google Glass and CityApp Are we fully aware of things happening around?

The lowest price for household appliance The traffic jam inside or around a city (EMBs are too sparse to satisfy the need) The weather of a particular place (weather forecast doesn’t really tell you that, does it?) “Making city-wide services accessible to every citizen”, according to SAP

A technological matchmaking

CityApp presents city-scale info Google Glass retrieves data with a multi-dimensional presentation

Jun Luo (NTU) Crowd Sensing Complexity Community 9 / 17

The Social Aspects of Crowd Sensing Challenges as A Social Behavior

Crowd Sensing: A Double-Side Sword Upside

Deliver multi-dimensional data as we have discussed, hence having the potential to “peep” deeper into the behavior of nature, as well as human societies

Downside

Operated by large populations has made crowd sensing itself a social behavior, hence being subject to the interference of human nature

Let us focus on the bad news first Human are selfish Human are unreliable

Jun Luo (NTU) Crowd Sensing Complexity Community 10 / 17

The Social Aspects of Crowd Sensing Challenges as A Social Behavior

Incentives and Quality Control for Crowd Sensing Why the crowd is willing to serve?

Previous applications are mostly based on reciprocity However, certain crowd sensing applications may not show reciprocity immediately, e.g., city-wide PSI monitoring Incentives (particularly monetary ones) are needed, but people are often not truthful: what if they claim more than needed?

Now we have less professional sensors and personnel

The volume of data is big but the data are noisy Data analytic technologies are needed to handle crowd sensed data Selecting right sensing personnel can be crucial, for which we need an

• nline learning approach

In both cases, we are under budget limit, and the applications may have other requires in terms of scheduling the individual sensing tasks!

Jun Luo (NTU) Crowd Sensing Complexity Community 11 / 17

The Social Aspects of Crowd Sensing Truthful Scheduling: A Theoretical Problem

Crowd Sensing vs. Crowd Sourcing The are not the same! Crowd Sourcing is mostly intellectual and bears no obligation, whereas Crowd Sensing are more physical: hence a social behavior that should be incentived We argue that time is one of the key difference Crowd Sensing

Participants’ available time for sensing is limited; the revenue gained from sensing is highly related to the length of sensing time

Crowd Sourcing

Not much related to time

Jun Luo (NTU) Crowd Sensing Complexity Community 12 / 17

The Social Aspects of Crowd Sensing Truthful Scheduling: A Theoretical Problem

Budget Limited Scheduling Task, Value, and Budget

A set of sensing tasks K = {K1, K2, · · · , Km}, a task Ki conducted per unit time has a sensing value ui. The owner of these tasks holds a budget G for recruiting the crowd A set participants A = {A1, A2, · · · , An}, Each Aj can perform one task at a time, has a private value dj for sensing cost per unit time, and is available during a time period [sj, ej]

Scheduling Crowd Sensing Bids ant Payments

Owner solicits bids bj = (dj, sj, ej) from participants, finds a schedule yj(b) ⊆ [sj, ej] for participant Aj, and computes a payment pj(b) for Aj

Jun Luo (NTU) Crowd Sensing Complexity Community 13 / 17

The Social Aspects of Crowd Sensing Truthful Scheduling: A Theoretical Problem

The Problem of Budget-Limited Truthful Scheduling Key objectives Owner maximizes the sensing revenue: Maximize R (y(b)) = m

i=1 ui

• j:kj=i yj(b)
• Budget feasibility: total payment is no more than G

Truthfulness: a participant maximizes its utility pj(b) − dj|yj(b)| by revealing its real private information, no matter how others act Individual Rationality: any truthful user gets non-negative utility The problem is novel to the best of our knowledge The traditional VCG mechanism is not applicable, due to the NP-hardness as well as the budget constraint Multi-parameter mechanism design is hard: each participant bids three parameters: di, si, and ei

Jun Luo (NTU) Crowd Sensing Complexity Community 14 / 17

The Social Aspects of Crowd Sensing Scheduling and Incentive Mechanism Design

A Three-Stage Design Methodology A divide-and-conquer approach An scheduling algorithm for handling the NP-hard side without considering truthfulness and rationality A specially designed payment rule extend the traditional“Myerson’s rule” to the multi-parameter case A “secretary algorithm” again for scheduling but only choosing one participant Combining these component in a randomized manner With probability 1/2 running the scheduling algorithm, and apply

• ur payment mechanism

With probability 1/2 running the “secretary algorithm”, and give the chosen one all budget

Jun Luo (NTU) Crowd Sensing Complexity Community 15 / 17

The Social Aspects of Crowd Sensing Scheduling and Incentive Mechanism Design

The Outcome and What Next Goals are achieved at a cost Our design achieves budget feasibility, truthfulness, individual rationality, and an approvable approximation ratio But the resulting payments for hiring participant are high in practice in order to thwart untruthfulness Some new targets What if the participants arrive in an online manner, rather than give as a set? Some more intriguing questions: how do people actually decide their private value? Would it be a constant for a certain person or a certain task? Is time playing a role again? How social evolution affects private values (we have seen similar things in crowd sourcing already)?

Jun Luo (NTU) Crowd Sensing Complexity Community 16 / 17

Conclusion

Back to Crowd Sensing In General – A Summary A few take-away points Crowd sensing widens our view, making us smarter, but it simply means we, as a whole, help ourselves One key is to change the viewpoint: crowd sensing may not acquire the exact information, but we can find other ways Leveraging the diversified sensing ability of devices around us and the scale of crowd sensed data helps us solving problems while bring forward new challenging problems Incentive and quality control schemes may need to be in place for applications that do not show immediate reciprocity For incentive mechanism design, pure mathematical reasoning may not solve problem in practice, but we are yet to understand the real complexity through more system developments

Thank you for your attention! Questions?

Jun Luo (NTU) Crowd Sensing Complexity Community 17 / 17