During Artificial Intelligence in Industry 4.0 Workshop, keynote speaker Michał Krasoń, a Data Scientist from Stermedia, shared his knowledge about Machine Learning from basic definitions to advanced case studies such as nitrous oxide emissions in turbines and anomaly detection from video streams.
Want to know more? Let's talk: hello@stermedia.eu or e-mail Michał directly: michal.krason@stermedia.eu
Breaking the Kubernetes Kill Chain: Host Path Mount
Python and Machine Learning Applications in Industry
1. Python and Machine Learning
Applications in Industry
Michał Krasoń
Data Scientist
AI Engineer
Stermedia.ai
2. INTELLIGENT
FACTORIES
AI DRIVEN
FACTORIES
By leveraging data, sensors, and
robots, U. S. manufacturing
output is 40% higher that two
decades ago.Due to the use of
data, sensors and robots,
industrial production is 40%
higher than 20 years ago
By incorporating advanced IoT
technologies and AI, factories can
generate trillions trillions by 2025
Inefficiencies and
waste result in billions
of dollars lost every
year
60 -70%
of wasted
materials
40%
higher output
$3.7
trillion
in new value
AI changes the production
TRADITIONAL
FACTORIES
3. why ML and AI is important now?
Key factors:
• Hardware
• Algorithms
• Data
• Networking
• Cloud
4. What is machine learning?
Artificial intelligence (AI)
Machine learning (ML)
Deep learning (DL)
Deals with processes in which the
computer solves tasks in a way that
imitates human behavior.
Artificial intelligence
(AI)
Machine learning (ML) Algorithms that allow computers to
learn from examples without being
explicitly programmed
Deep learning (DL) An ML subset that uses deep neural
networks as models
5. ML vs non-ML
‘Traditional’ programming
Machine learning
Output
Historical
output
New
data
Historical
data
Data
Program
(model)
Program
Output
16. deep learning
1. A lot of complex data (texts, images)
2. Neural networks
a. "Classic" deep networks
b. Convolutional - image recognition (CV)
c. Recurrent - Natural Language Processing (NLP)
17. why Python?
Less code to achieve results
A lot of ML-dedicated libraries
Readable, structured code
Easy prototyping
Production - ready
19. cs: prediction of demand
Situation:
production hall, we want to predict the use of additional parts on individual days
Data
the need for elements and consumption of additional parts in the last 4 months, the need for elements in the future
Problem:
few dates, many elements affecting consumption
20. krok 1: wybór klasy modeli, train- test split
1. Too little data to use advanced ML algorithms
2. Dependence is intuitively linear - we use linear models
The teaching part The test part
21. step 2: limiting the number of variables
Too many elements make modeling difficult. We do feature selection:
1. Study of correlation with the target variable
2. Dimensionality reduction, grouping of elements
3. Use of automatic methods of eliminating features
22. step 3: choice of model, prediction
1. We adjust models to learning data, we choose the best based on cross validation
2. We perform the prediction on the test set
23. Michał Krasoń
Data Scientist
e-mail: michal.krason@stermedia.ai
Linkedin: https://www.linkedin.com/in/michalkrason
www.stermedia.ai
Thank you :)