The Role of Taxonomy and Ontology in Semantic Layers - Heather Hedden.pdf
FIWARE Global Summit - Advanced ML/AI Techniques with FIWARE and Connected IoT Devices
1. Scale Up for a Real Smart Future
Berlin, Germany
23-24 October, 2019
Advanced ML/AI Techniques with FIWARE
and Connected IoT Devices
Adrián Arroyo
IoE Lab. Research & Innovation, Atos Spain, Spain
adrian.arroyo@atos.net
@arroyadr
2. How to address the challenge of transforming data
into intelligence?
§ IoT ↔ Big Data
§ AI technologies and Machine Learning techniques are
suitable for IoT scenarios
• Detecting patterns and behaviours from gathered data
§ Analysing these patterns, it is possible to transform data
into intelligence
• Generating lots of smart services
3. How to address the challenge of transforming data
into intelligence?
§ Appearance of new IoT hardware with better performance
capabilities on AI in the Edge
§ Power demanding is an issue for IoT
• Hardware performance
• Connectivity issues
• Energy efficiency
§ Move data and intelligence to the cloud is not a solution for
IoT
• Accessibility to the cloud is not guaranteed every time
(24h/7d)
• Delays in responses may appear
5. EASIER: Architecture
§ Data manager: brings IoT data to the
platform (NGSIv2 supported)
§ Automatic Cloud/Edge data
synchronization
§ Cloud platform to implement your data-
science tasks: create, train and test
models with the Micro Trainers
§ Use the models in the edge (or cloud) with
automatic model synchronization with
the Micro inferencers
§ Models optimized for specific hardware
platforms (TPU, Raspberry Pi, etc.).
12. Use Case:
§ https://synchronicity-iot.eu/
§ Data ingestion (NGSIv2): Orion, STH
Comet, …
§ Data sources: Parking, Traffic, Noise
§ Smart services
• Prediction of future data
Data
Manager
• Free/occupied ratio vs time of day