Machine Learning orientation
1
Introduction
I Machine learning: Shall we?
1.1
What is intelligence?
1.1.1
Definition of artificial intelligence sub domains
1.2
Is AI smarter than humans?
1.2.1
Thinking, fast and slow
(Kahneman
2011
)
1.3
Comparisons between AI and humans
1.3.1
Breast cancer detection
1.3.2
Working together: Lung cancer detection
1.3.3
ImageNet Large Scale Visual Recognition Challenge (ILSVRC)
1.3.4
AlphaGo Zero
1.4
ML models with bias
1.5
Attacks on ML models
1.5.1
Adding noise to image leads to misclassification
1.5.2
But what about attacks on human perception?
1.5.3
Model Hacking ADAS to Pave Safer Roads for Autonomous Vehicles
1.6
Measuring the Algorithmic Efficiency of Neural Networks TBD
2
Data Ethics
2.1
Topics in data ethics
2.1.1
Recourse and accountabilty
2.1.2
Feedback loops
2.1.3
Bias
2.2
Identify and adressing ethical issues
2.2.1
Analyze a planned project
2.2.2
Process to implement
2.2.3
Diversity
2.3
Data Ethics consequences
2.3.1
Researcher stops his work due to ethical concerns
2.3.2
Career: Oxford seeks AI ethics professor
2.3.3
Shaping Europe’s digital future: Commission presents strategies for data and Artificial Intelligence
3
Strategies for machine learning
3.1
Management ML strategy: 7 steps for a successful ML project
3.2
Management ML strategy: Data Project Checklist
3.3
Project management ML strategy: The Drivetrain Approach
3.3.1
Recommendation system
3.3.2
Exercise: Optimizing lifetime customer value
3.4
Developer ML strategy TBD
4
Outlook
4.1
Development of life
4.1.1
When will superhuman AI come, and will it be good?
4.1.2
AI aftermath scenario
4.2
Data religion: Dataism
II Machine learning fundamentals
4.3
Identify if ML is suited to fulfill need
4.4
Gather data
4.4.1
What kind of data can be used?
4.4.2
How much data is necessary?
4.4.3
Which data is useful?
4.5
Exploratory and quantitative data analysis
4.5.1
Example for exploratory and quantitative data analysis
4.5.2
Visualizations for Categorical Data: Exploring the OkCupid Data
4.6
Feature engineering
4.6.1
Encoding Categorical Predictors
4.6.2
Engineering numeric features
4.6.3
Feature importance
4.7
Model fit
4.8
Model tuning
4.8.1
Metrics
4.9
Deploy model
4.10
Supervised learning
4.10.1
Self supervised learning
4.11
Unsupervised learning
4.11.1
Discovering clusters
4.11.2
Discovering latent factors
4.12
Reinforcement learning
4.12.1
Elements of reinforcement learning
4.12.2
RL algorithms
4.12.3
Example self driving car MIT
5
ML algorithms
5.1
Linear regression
5.1.1
Example for linear regression
5.2
Logistic regression
5.2.1
Python example logistic regression
5.3
Tree based methods
5.3.1
Splitting metrics
5.3.2
Ensembles
5.3.3
Random forest
5.3.4
Boosted trees
5.4
Support Vector Machine (SVM) TBD
5.4.1
Kernels
5.4.2
Python example for SVM
5.5
Neural networks
5.5.1
Geometric Intuition for Training Neural Networks
5.5.2
Convolutional Neural Network (CNN) TBD
5.5.3
RNN TBD
5.5.4
GANs
5.6
A Gentle Introduction to CycleGAN for Image Translation
5.6.1
Examples for GANs
5.7
Software that can generate photos from paintings, turn horses into zebras, perform style transfer, and more.
5.7.1
Pix2pix framework
5.8
Transformers TBD
5.8.1
Example transformer for time series forecasting
5.9
NLP
5.10
Dataset search from Google
6
ML application examples
6.1
AI translation by DeepL
6.2
Artificial intelligence detects myocardial infarctions in the ECG more reliably than cardiologistsn
6.2.1
First in Germany: Artificial intelligence recognizes COVID-19 in clinical routine
6.3
A Deep Learning Approach to Antibiotic Discovery
6.4
Fundamental limits from chaos on instability time predictions in compact planetary systems
6.4.1
Comparison between SPOCK and previous models
6.5
Machine Learning Algorithms and Global Optimization Methods for SPICE Model Parameter Extraction
6.6
How many yards will an NFL player gain after receiving a handoff?
6.7
Predictive Maintenance for the elevator and escalator industry TBD
6.8
Disease outbreak risk software
6.9
Neural networks enable autonomous navigation of catheters
6.10
Bosch FLEXIDOME IP starlight 8000i
6.11
Master Autonomous Driving
6.12
University Suttgart: Indoor-Ortung mit Mobilfunk
III Explainable machine learning
6.13
Method: Layer-Wise Relevance Propagation
6.14
Method: SpRay
6.15
Method: Salient-object-visualization
6.15.1
Tool: keras-salient-object-visualization
6.15.2
Explaining How a Deep Neural Network Trained with End-to-End Learning Steers a Car
6.16
Tool: Lime
6.17
Tool: tf-explain tbd
6.18
Tool: alibi
IV ML online resources
6.18.1
Example for self-driving car course project
6.19
fast.ai
6.19.1
fast.ai Book
6.20
Full Stack Deep Learning
6.21
In-depth introduction to machine learning in 15 hours of expert videos
6.21.1
An Introduction to Statistical Learning
6.22
The learning machine
6.23
DeepAI: The front page of A.I.
6.24
TensorFlow tutorials
6.24.1
MIT 6.S191 Introduction to Deep Learning
6.25
Embedding Projector
6.26
Tensorboard playground
6.27
Empowering companies to jumpstart AI and generate real-world value
6.28
TensorFlow, Keras and deep learning, without a PhD
6.29
Neural Networks and Deep Learning
6.30
Platform.ai: produce high-quality labels
6.31
Neural Networks and Deep Learning
6.32
Deep Learning
V Examples from Kaggle
6.33
Winning solution (1st)
6.33.1
Alex / Gilberto models
6.33.2
Feng models
6.33.3
Andriy models
6.33.4
Code on GitHub
6.33.5
Input features and embeddings
6.33.6
Ensembling
6.33.7
Hardware
6.33.8
Software
6.33.9
Code on GitHub
6.34
#2 solution 🤖 Quantum Uncertainty 🤖
6.34.1
Overall architecture
6.34.2
Input features and embeddings
6.34.3
Data augmentation
6.34.4
Ensembling
6.34.5
Hardware
6.34.6
Software
6.34.7
Code on GitHub
VI Real world example
7
Real world example
7.1
Subject of the project
Depending from where you were looking:
Looking from the perspective of machine learning expert
7.2
Project phases
The main project phases are:
After data gathering iteration is trump
7.2.1
Feature engineering
7.3
Algorithm selection
7.3.1
Logistic regression
7.3.2
Tree based
7.3.3
Support Vector Machine (SVM) TBD
7.4
Performance measurement
7.4.1
Sensitivity and specificity
7.4.2
Receiver operating characteristic (ROC)
7.5
Confusion matrix and receiver operating characterstic (ROC) for pulse
7.5.1
Receiver operating characterstic (ROC) and probability density plots
7.6
Create augmented labeled data
7.6.1
Features of time signals
7.7
Features generated
7.7.1
Analysis of generated features
7.7.2
Dynamic time warp (DTW) for signal
7.8
Confusion matrix results logistic regression for measured data
7.8.1
ROC results for measured data
7.9
Several algorithms results for SNR = 18dB
7.9.1
ROC results for SNR 18dB
7.10
Calculation of return of invest (ROI)
7.10.1
Calculation of ML project invest
7.10.2
Calculation of ML profit
7.10.3
Resulting ROI
7.11
Compare models for SNR = 18dB
7.12
Optimize ML hyper parameter
VII Cloud-based machine learning
8
Cloud-based machine learning
VIII Kaggle Survey
8.1
Kaggle survey details
8.2
Purpose
8.3
Navigation and handling
8.4
Survey participants education level
8.5
Who uses which algorithm
8.6
Machine learning experience and algorithms
8.7
Experience and new algorithms
8.8
Role of participants
8.9
Company size
8.10
Company incorporation of machine learning
8.11
Favourite media sources on data science topics
8.12
Favourite online course platform
8.13
Favourite data analyzing tool
8.14
Experience in data analysis coding
8.15
Favourite integrated development environments (IDE’s)
8.16
Favourite hosted notebook products
8.17
Favourite programming languages
8.18
Recommended entry programming language
8.19
Favourite data visualization libraries or tools
8.20
Favourite specialized hardware
8.21
Favourite machine learning frameworks
8.22
Favourite cloud computing platforms
8.23
Favourite big data / analytics products
8.24
Favourite automated machine learning tools (or partial AutoML tools)
References
Published with bookdown
Machine learning orientation
7.12
Optimize ML hyper parameter
