1. High-integrity Sensor Networks Mani Srivastava mbs@ucla.edu UCLA

2. High Integrity Sensor Networks Information returned may be distorted Malicious attack (privacy, denial of service, corruption) Noise: random error, unavoidable In the sensor In communication Fault: Persistent, transient, intermittent malfunction Easy to counter Hard to learn Hard to counter Easy to learn NEW FOCUS

3. Key Issues and Challenges Sensor nodes have to be cheap –Fault-tolerant, secure hardware not feasible Unmanned and untethered –No central supervision Severe resource constraints –Limited memory, CPU, b/w Scale Collaborative behavior –Decentralized in-network processing for energy efficiency Open wireless medium Handle misbehaving nodes Existing solutions not portable End-to-End Model? Robust Relaying

4. Spectrum of Integrity Problems Misbehavior while routing event report Well-studied ad hoc routing problems I detected a car at (x,y) I detected a car at (x,y) I detected a car at (x,y) Misbehaving Node

5. Spectrum of Integrity Problems It was a false alarm. No such car! I detected a car at (x,y) I detected a car at (x,y) Misbehavior during in-network processing (e.g. generation of bad event, suppression of event) Well-studied fault tolerant computing problems Misbehaving Node

6. Spectrum of Integrity Problems No car Detected I detected a car at (x,y) No car Detected Misbehavior during information generation (e.g. miscalibrated or compromised sensors) Our focus!

7. Approach: Reputation-based Framework Key Question: How do nodes trust each other? Embedded in every social network is a web of trust How do human societies evolve? –Principle of reciprocal altruism: Be nice to others who are nice to you –When faced with uncertainties: Trust those who have reputation of being trustworthy Our Approach: Form a similar community of trustworthy nodes in the network over time Why? –Sensor networks already follow a community model –Successful centralized instantiation in networks such as E-Bay But missing element is trust as nodes are dumb –Trust even nodes that are fault or compromised by internal adversaries

8. Basic Concept - Learning from Experience n Observe the action of other nodes – Watchdog mechanism n Develop a perception of other nodes over time – Reputation n Predict their future behavior – Trust n Cooperate/Non-cooperate with trustworthy nodes – Behavior n Share experiences to facilitate community growth – Second hand info Watchdog mechanism Reputation TrustBehavior Second hand information

9. Skeleton Structure of RFSN Algorithm Development Monitoring, Data Analysis Statistics…. Networking, Cryptography Development of high integrity sensor networks will be a combination of techniques from different fields Watchdog mechanism Reputation TrustBehavior Second hand information Decision theory Bayesian Analysis…

10. Bayesian Formulation of Sensor Integrity Reputation representation for neighbors as beta distribution –two parameters (cooperation, non-cooperation) Easy to compute reputation updates and integration using belief theory –Updates –Integration Reputation propagation via public key crypto for node authentication Trust to estimate future cooperative behavior  of node j with reputation Beta(  j,  j ) Classifying node behavior as good/bad by simple threshold on trust New reputation, ?? Old reputation, Beta(  j,  j ) ????

11. Conclusions Reputation-based approach is –Scalable: localized algorithms –General: communication, sensing, processing –Unified: faulty, malicious, noisy But insufficient for certain malicious attacks on integrity –Context-aware adversaries –Resource consumption attacks –Compromises of the physical world Network level techniques? Multimodality? No Event