Gossip-based Truth Discovery Motivation Motivating Examples - PowerPoint PPT Presentation

Gossip-based Truth Discovery Zhiying Xu Introduction Gossip-based Truth Discovery Motivation Motivating Examples Problem Formulation Zhiying Xu Computational Complexity Proposed Algorithms Exact June 4, 2017 Algorithm Approximation Algorithms Decentralize and Randomize Decentralized Algorithms Speed Up . . . . . . . . . . . . . . . . . . . . Gossip . . . . . . . . . . . . . . . . . . . . Algorithm 1 / 36 Future Work

Outline Gossip-based 1 Introduction Truth Discovery Motivation Zhiying Xu Motivating Examples Introduction 2 Problem Formulation Motivation Motivating Examples 3 Computational Complexity Problem Formulation 4 Proposed Algorithms Computational Exact Algorithm Complexity Approximation Algorithms Proposed Algorithms Exact 5 Decentralize and Randomize Algorithm Approximation Decentralized Algorithms Algorithms Decentralize Speed Up Gossip Algorithm and Randomize 6 Future Work Decentralized Algorithms Speed Up . . . . . . . . . . . . . . . . . . . . Gossip . . . . . . . . . . . . . . . . . . . . Algorithm 2 / 36 Future Work

Outline Gossip-based 1 Introduction Truth Discovery Motivation Zhiying Xu Motivating Examples Introduction 2 Problem Formulation Motivation Motivating Examples 3 Computational Complexity Problem Formulation 4 Proposed Algorithms Computational Exact Algorithm Complexity Approximation Algorithms Proposed Algorithms Exact 5 Decentralize and Randomize Algorithm Approximation Decentralized Algorithms Algorithms Decentralize Speed Up Gossip Algorithm and Randomize 6 Future Work Decentralized Algorithms Speed Up . . . . . . . . . . . . . . . . . . . . Gossip . . . . . . . . . . . . . . . . . . . . Algorithm 3 / 36 Future Work

Outline Gossip-based 1 Introduction Truth Discovery Motivation Zhiying Xu Motivating Examples Introduction 2 Problem Formulation Motivation Motivating Examples 3 Computational Complexity Problem Formulation 4 Proposed Algorithms Computational Exact Algorithm Complexity Approximation Algorithms Proposed Algorithms Exact 5 Decentralize and Randomize Algorithm Approximation Decentralized Algorithms Algorithms Decentralize Speed Up Gossip Algorithm and Randomize 6 Future Work Decentralized Algorithms Speed Up . . . . . . . . . . . . . . . . . . . . Gossip . . . . . . . . . . . . . . . . . . . . Algorithm 4 / 36 Future Work

Gossip Algorithm Gossip-based Gossip algorithms : schemes which distribute the computation Truth Discovery burden and in which a node communicates with a randomly Zhiying Xu chosen neighbor. Introduction Motivation Motivating Examples Problem Formulation Computational Complexity Proposed Algorithms Exact Algorithm Approximation Algorithms Decentralize and Randomize Decentralized Algorithms Speed Up . . . . . . . . . . . . . . . . . . . . Gossip . . . . . . . . . . . . . . . . . . . . Algorithm 5 / 36 Future Work

Truth Discovery Gossip-based Truth Discovery Zhiying Xu Introduction Motivation Motivating Examples Discover truth from noisy crowd sensing data Problem Formulation Evaluate the reliability of sensors Computational Complexity Proposed Algorithms Exact Algorithm Approximation Algorithms Decentralize and Randomize Decentralized Algorithms Speed Up . . . . . . . . . . . . . . . . . . . . Gossip . . . . . . . . . . . . . . . . . . . . Algorithm 6 / 36 Future Work

Outline Gossip-based 1 Introduction Truth Discovery Motivation Zhiying Xu Motivating Examples Introduction 2 Problem Formulation Motivation Motivating Examples 3 Computational Complexity Problem Formulation 4 Proposed Algorithms Computational Exact Algorithm Complexity Approximation Algorithms Proposed Algorithms Exact 5 Decentralize and Randomize Algorithm Approximation Decentralized Algorithms Algorithms Decentralize Speed Up Gossip Algorithm and Randomize 6 Future Work Decentralized Algorithms Speed Up . . . . . . . . . . . . . . . . . . . . Gossip . . . . . . . . . . . . . . . . . . . . Algorithm 7 / 36 Future Work

A Real-life Case Gossip-based Whether the class is canceled? Truth Discovery Zhiying Xu Introduction Motivation Motivating Examples Problem Formulation Computational Complexity Proposed Algorithms Exact Algorithm Approximation Algorithms Decentralize and Randomize Decentralized Algorithms Speed Up . . . . . . . . . . . . . . . . . . . . Gossip . . . . . . . . . . . . . . . . . . . . Algorithm 8 / 36 Future Work

Outline Gossip-based 1 Introduction Truth Discovery Motivation Zhiying Xu Motivating Examples Introduction 2 Problem Formulation Motivation Motivating Examples 3 Computational Complexity Problem Formulation 4 Proposed Algorithms Computational Exact Algorithm Complexity Approximation Algorithms Proposed Algorithms Exact 5 Decentralize and Randomize Algorithm Approximation Decentralized Algorithms Algorithms Decentralize Speed Up Gossip Algorithm and Randomize 6 Future Work Decentralized Algorithms Speed Up . . . . . . . . . . . . . . . . . . . . Gossip . . . . . . . . . . . . . . . . . . . . Algorithm 9 / 36 Future Work

Model Gossip-based Truth Discovery Definition (Problem Formulation) Zhiying Xu Given a crowd sensing model where N sensors make individual observations X N × M ∈ { 0 , 1 } N × M about a set of M variables, Introduction Motivation determine the true value of these variables t M ∈ { 0 , 1 } M which Motivating Examples satisfies Problem Formulation ⟨ t , r ⟩ = arg min p ( X | t , r ) , Computational ⟨ t , r ⟩ Complexity where r N ∈ [ 0 , 1 ] is the reliability of sensors. Proposed Algorithms Exact Algorithm Approximation Algorithms Decentralize and Randomize Decentralized Algorithms Speed Up . . . . . . . . . . . . . . . . . . . . Gossip . . . . . . . . . . . . . . . . . . . . Algorithm 10 / 36 Future Work

Model Gossip-based Truth Discovery Definition (Problem Formulation) Zhiying Xu Given a crowd sensing model where N sensors make individual observations X N × M ∈ { 0 , 1 } N × M about a set of M variables, Introduction Motivation determine the true value of these variables t M ∈ { 0 , 1 } M which Motivating Examples satisfies Problem Formulation ⟨ t , r ⟩ = arg min p ( X | t , r ) , Computational ⟨ t , r ⟩ Complexity where r N ∈ [ 0 , 1 ] is the reliability of sensors. Proposed Algorithms Exact Algorithm Approximation Algorithms t , r are related to each other. Decentralize There is no need for iteration by EM algorithm. and Randomize Decentralized Algorithms Speed Up . . . . . . . . . . . . . . . . . . . . Gossip . . . . . . . . . . . . . . . . . . . . Algorithm 10 / 36 Future Work

Objective Function Gossip-based Truth Discovery Zhiying Xu Using the relationship between t and r , We can simplify the Introduction previous problem into the optimization of the objective function Motivation Motivating Examples N Problem ∑ t = arg min f ( x i , t ) , Formulation t Computational i = 1 Complexity where f ( x i , t ) denotes Proposed Algorithms ( ) ( ) || x i − t || 1 − || x i − t || M M Exact ln || x i − t || + ln 1 − . Algorithm M M || x i − t || Approximation Algorithms Decentralize and Randomize Decentralized Algorithms Speed Up . . . . . . . . . . . . . . . . . . . . Gossip . . . . . . . . . . . . . . . . . . . . Algorithm 11 / 36 Future Work

Truth Discovery Strategy Gossip-based Truth Discovery Zhiying Xu Introduction Motivation All sensors are in a connected networks. Motivating Examples Sensors have limited memory and can’t store much data. Problem Formulation Each sensor communicates with a random neighbor Computational Complexity asynchronously. Proposed Algorithms Exact Algorithm Approximation Algorithms Decentralize and Randomize Decentralized Algorithms Speed Up . . . . . . . . . . . . . . . . . . . . Gossip . . . . . . . . . . . . . . . . . . . . Algorithm 12 / 36 Future Work

Outline Gossip-based 1 Introduction Truth Discovery Motivation Zhiying Xu Motivating Examples Introduction 2 Problem Formulation Motivation Motivating Examples 3 Computational Complexity Problem Formulation 4 Proposed Algorithms Computational Exact Algorithm Complexity Approximation Algorithms Proposed Algorithms Exact 5 Decentralize and Randomize Algorithm Approximation Decentralized Algorithms Algorithms Decentralize Speed Up Gossip Algorithm and Randomize 6 Future Work Decentralized Algorithms Speed Up . . . . . . . . . . . . . . . . . . . . Gossip . . . . . . . . . . . . . . . . . . . . Algorithm 13 / 36 Future Work