ADOS: An advanced AI-based oracle system for securing off-chain IoT data integrity when injecting in the blockchain

Slides presented at the ONTOCHAIN Summit for Trustworthy Internet by Juan Miguel Navarro, Professor at UCAM university & ADOS and AirTrace technical coordinator

Transcript

1. CLICK TO EDIT MASTER TITLE STYLE Click to add subtitle Location Date ADOS (AIRTRACE DECENTRALIZED ORACLE SYSTEM) An advanced AI-based oracle system for securing off-chain IoT data integrity when injecting in the blockchain OntoChain Summit 22 Berlin, 2 June ® ANY IOT SENSOR, BLOCKCHAIN-ENABLED
2. | ONTOCHAIN.NGI.EU 2 Visual: Web user interface for fast integration Any IoT sensor, blockchain-enabled (Learn more at https://airtrace.io) ®
3. | ONTOCHAIN.NGI.EU 3 Programmatic: RESTFUL API and IoT protocols Any IoT sensor, blockchain-enabled (Learn more at https://airtrace.io) ®
4. | ONTOCHAIN.NGI.EU 4 WHY ANOMALY DETECTION IN BLOCKCHAIN -IOT DOMAINS?
5. | ONTOCHAIN.NGI.EU 1.-Machine Learning algorithm with distributed computation 2.- Ontology for IoT-ML models 3.- API for integration with 3rd parties Contributions of the project
6. | ONTOCHAIN.NGI.EU 6 ML MODELS FOR DEEPER UNDERSTANDING Three types of anomalies Point: irregularity that happens randomly and may have no particular reason. Contextual: abnormal behavior happening within some specific context. Collective: a collection of individual data points showing anomalies.
7. | ONTOCHAIN.NGI.EU 7 SUPPORTING ARTIFICIAL INTELLIGENCE IN IOT FOR EARLY ANOMALY DETECTION, TIMELY ALERTING, AND DATA- QUALITY SCORING, BEFORE INJECTING INTO THE BLOCKCHAIN ADOS - GNN Deng, A., & Hooi, B. (2021, February). Graph neural network-based anomaly detection in multivariate time series. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 35, No. 5, pp. 4027-4035).
8. | ONTOCHAIN.NGI.EU 8 ARCHITECTURE OF THE SOLUTION Client n Client 1 ADOSTM IoT System Integrators Model weights S3/IPFS Periodic model training Sensor readings Worker 2 Worker 1 Worker 3 Worker 4 Worker n Worker 5 Worker 6 Data quality factors Middleware (WoT) Blockchain
9. | ONTOCHAIN.NGI.EU 9 USE STORIES AND USE-CASE Different scenarios:  Water quality sensors: in water treatment plants, it is possible to find several sensors measuring water levels, water quality, valve status, flow rates, etc. Data measured by these IoT sensors can be related in complex, nonlinear ways.  Acoustic sensors: in smart cities, acoustic changes can reveal changes in car traffic, weather conditions, etc.. Acoustic sensors are used to monitor the sound pressure level (noise pollution) of the city.  Transportation sector: to understand the existing dynamics arising in scenarios of car traffic in order to understand how vehicular mobility of several vehicles can affect each other .
10. 10 ONTOLOGICAL RESOLUTION MIDDLEWARE (WOT) Name Description Domain New/MyUpdated/ MyReused/Reused Language AOM (ADOS Ontology Model) Regulates data for anomaly detection models to work properly (e.g. model version, TEE needed, etc.) Machine Learning model specification New WoT SSN Sensors and their observations, the involved procedures, and the observed properties. Systems (sensors and actuators) Reused WoT WGS84 Represents geocoordinates (latitude and longitude), with class geo:point and properties geo:latitude, geo:longitude, geo:altitude. Geolocation (outdoors) Reused WoT DUL By combining WSG84 properties with dul:hasLocation from DUL ontology we can represent the sensor location indoors Geolocation (indoors) Reused WoT FIEMSER Represents the way to communicate with the IoT devices, including, at least, communication protocol and its version. Communications Reused WoT QUDT Represents a comprehensive list of quantities, units and dimensions to define IoT readings formats. Observations Reused WoT FOAF Information about IoT vendor and version of the device Vendor, version and deployment time Reused WoT
11. | ONTOCHAIN.NGI.EU 11 DEMO LINK
12. - Training models with distributed computation - Graph tools for anomaly interpretation and analysis - Exploration of other ML techniques Future extensions
13. THANKS FOR YOUR SUPPORT QUESTIONS??