Artificial intelligence has become accessible on a wide range of platforms, and may prove indispensible to extract valuable information and insights from the massive amounts of sensor data - both structured and unstructured- that are available and will grow to brontobyte levels in the coming years.
1. IoT won’t work without Artificial Intelligence
IoT Convention 2018 – Antwerp
2. Speaker bio
Education
Engineer’s degree, Civil Engineer
in Computer science
Professional experience
IBBT/iMinds : 2004-2007
LoQutus : 2007 – current
Certification
Fundamentals of Enterprise
architecture
Archimate 2.0
IBM Certified SOA Associate
CTO & Lead strategist Enterprise integration solutions
Experienced solution and enterprise architect with a very keen interest in digital technologies, and how
these technologies can be used to optimize and disrupt business models and ecosystems :
• Cloud/hybrid integration, microservices andAPI Management
• Mobile technologies & Social media integration
• Analytics and Big data
• Internet-of-Things
• Blockchain and digital ledger technology
vincent.verstraete@loqutus.com
https://www.linkedin.com/in/vvstraet/
9. General programming vs Machine Learning
Machine LearningGeneral programming
INPUT
PROGRAM
INPUT
OUTPUT
OUTPUT
Learned
Program
NEW INPUT INFERRED OUTPUT
10. Artificial intelligence over time
1960s – 1980s 1980s – 2020s ???
AI Research
Initial AI concepts
Machine learning
“Narrow AI”
General AI
Applies intelligence to
any problem
Deep learning
FOCUS
13. Machine learning process
Follow-up on quality
Iterative improvement
Human interpretation
Improve
ML Model
(improved)
~20% of the data
Model scoring
Test
ML Model
(validated)
~80% of the data
Parameter tuning
Train model
ML Model
Data cleansing
Data
transformation
Data enrichment
Feature
engineering
Prepare data
Data set (clean)
Sensors
Historical data
Open data
Labeling
Data set (raw)
Find data sources
14. Takeaways
• Available services in all major cloud platforms
(Microsoft, Amazon, IBM, Google)
• Integrated in major application suites
(Salesforce, SAP, Excel, …)
• Low-threshold services
(but this does not mean they’re cheap)
Machine Learning tech is mature
Privacy & regulations
• AI services require (large amounts of) data input,
meaning they will see your data
• GDPR, consent, …
• Resistance is expected (job evolution)
Data is the starting point !
• The quality of the model is directly proportional to the
amount of training data.
• Data preparation (cleansing, …) is crucial.
• Changing the model requires a full regression test.
You can only predict the predictable
• “No one has learned to beat the stock market”
• Keep in mind that Machine Learning models don’t
“understand” what they are doing.