Machine Learning orientation
1
Introduction
I Machine learning: Shall we?
2
What is machine learning?
2.1
What is intelligence?
2.1.1
Definition of artificial intelligence sub domains
2.2
Is AI smarter than humans?
2.2.1
Thinking, fast and slow
2.2.2
Predictions
2.2.3
Decisions
2.3
Comparisons between AI and humans
2.3.1
Breast cancer detection
2.3.2
Working together: Lung cancer detection
2.3.3
ImageNet Large Scale Visual Recognition Challenge (ILSVRC)
2.3.4
AlphaGo Zero
2.4
ML models with bias
2.4.1
What is the relation between bias in machine learning and priors of Bayes theorem?
2.4.2
How to avoid bias
2.4.3
Human bias
2.5
Attacks on ML models
2.5.1
Adding noise to image leads to misclassification
2.5.2
But what about attacks on human perception?
2.5.3
Model Hacking ADAS to Pave Safer Roads for Autonomous Vehicles
2.6
Measuring the Algorithmic Efficiency of Neural Networks TBD
3
ML application examples
3.1
AlphaFold: a solution to a 50-year-old grand challenge in biology
3.1.1
Results of 2020 CASP
3.1.2
Why does it matter?
3.1.3
CASP details
3.2
AI translation by DeepL
3.3
Artificial intelligence detects myocardial infarctions in the ECG more reliably than cardiologistsn
3.3.1
First in Germany: Artificial intelligence recognizes COVID-19 in clinical routine
3.4
A Deep Learning Approach to Antibiotic Discovery
3.5
Fundamental limits from chaos on instability time predictions in compact planetary systems
3.5.1
Comparison between SPOCK and previous models
3.6
Solution of the Schrödinger equation using deep neural networks
3.7
Global Optimization Methods for SPICE Model Parameter Extraction
3.8
Fraud detection based on mouse movement images
3.9
How many yards will an NFL player gain after receiving a handoff?
3.10
Predictive Maintenance for the elevator and escalator industry TBD
3.11
BBC: Artificial intelligence-created medicine to be used on humans for first time
3.12
Disease outbreak risk software
3.13
Neural networks enable autonomous navigation of catheters
3.14
Bosch FLEXIDOME IP starlight 8000i
3.15
Demonstration of computer vision system “thumbs up”
3.15.1
Run demonstration on Jetson Nano
3.16
Master Autonomous Driving
3.17
University Suttgart: Indoor-Ortung mit Mobilfunk
4
Data Ethics
4.1
Topics in data ethics
4.1.1
Recourse and accountabilty
4.1.2
Feedback loops
4.1.3
Bias
4.2
Identify and adressing ethical issues
4.2.1
Analyze a planned project
4.2.2
Process to implement
4.2.3
Diversity
4.3
Data Ethics consequences
4.3.1
Researcher stops his work due to ethical concerns
4.3.2
Career: Oxford seeks AI ethics professor
4.3.3
Shaping Europe’s digital future: Commission presents strategies for data and Artificial Intelligence
5
Strategies for machine learning
5.1
The HOOD-ST2C ML strategy
5.2
Management ML strategy: 7 steps for a successful ML project
5.3
Management ML strategy: Data Project Checklist
5.4
Project management ML strategy: The Drivetrain Approach
5.4.1
Recommendation system
5.4.2
Exercise: Optimizing lifetime customer value
5.5
Developer ML strategy TBD
6
Outlook of ML future
6.1
Development of life
6.1.1
When will superhuman AI come, and will it be good?
6.1.2
AI aftermath scenario
6.2
Data religion: Dataism
II Machine learning fundamentals
7
ML project process
7.1
Identify if ML is suited to fulfill need
7.2
Gather data
7.2.1
What kind of data can be used?
7.2.2
How much data is necessary?
7.2.3
Which data is useful?
7.3
Data analysis
7.3.1
Example for exploratory and quantitative data analysis
7.3.2
Visualizations for Categorical Data: Exploring the OkCupid Data
7.4
Feature engineering
7.4.1
Encoding Categorical Predictors
7.4.2
Engineering numeric features
7.4.3
Feature importance
7.5
Model fit
7.6
Model tuning
7.6.1
Metrics
7.7
Model deployment
8
Machine learning types
8.0.1
Machine learning types by data type
8.0.2
Machine learning types by data content
8.1
Supervised learning
8.1.1
Self supervised learning
8.2
Unsupervised learning
8.2.1
Discovering clusters
8.2.2
Discovering latent factors
8.3
Reinforcement learning
8.3.1
Elements of reinforcement learning
8.3.2
RL algorithms
8.3.3
Example self driving car MIT
9
ML algorithms
9.1
Linear regression
9.1.1
Example for linear regression
9.2
Logistic regression
9.2.1
Python example logistic regression
9.3
Tree based methods
9.3.1
Splitting metrics
9.3.2
Ensembles
9.3.3
Random forest
9.3.4
Boosted trees
9.4
Support Vector Machine (SVM) TBD
9.4.1
Kernels
9.4.2
Python example for SVM
9.5
Neural networks
9.5.1
Geometric Intuition for Training Neural Networks
9.5.2
Convolutional Neural Network (CNN) TBD
9.5.3
RNN TBD
9.5.4
GANs
9.6
A Gentle Introduction to CycleGAN for Image Translation
9.6.1
Examples for GANs
9.7
Software that can generate photos from paintings, turn horses into zebras, perform style transfer, and more.
9.7.1
Pix2pix framework
9.8
Transformers TBD
9.8.1
Example transformer for time series forecasting
9.9
Naive Bayes Classifier
9.9.1
Gaussian Naive Bayes¶
10
Food for the algorithms: Data
10.1
NLP
10.2
Dataset search from Google
III Explainable machine learning
11
Explainable machine learning
11.1
Method: Layer-Wise Relevance Propagation
11.2
Method: SpRay
11.3
Method: Salient-object-visualization
11.3.1
Tool: keras-salient-object-visualization
11.3.2
Explaining How a Deep Neural Network Trained with End-to-End Learning Steers a Car
11.4
Tool: Lime
11.5
Tool: tf-explain
11.6
Tool: alibi TBD
IV ML online resources
12
ML online courses
12.1
Coursera
12.2
Udemy
12.3
DataCamp
12.4
Udacity
12.4.1
Example for self-driving car course project
12.5
fast.ai
12.5.1
fast.ai Book
12.6
Kaggle Courses
12.7
Full Stack Deep Learning
13
ML online resources
13.1
In-depth introduction to machine learning in 15 hours of expert videos
13.1.1
An Introduction to Statistical Learning
13.2
The learning machine
13.3
DeepAI: The front page of A.I.
13.4
TensorFlow tutorials
13.4.1
MIT 6.S191 Introduction to Deep Learning
13.5
Embedding Projector
13.6
Tensorboard playground
13.7
Empowering companies to jumpstart AI and generate real-world value
13.8
TensorFlow, Keras and deep learning, without a PhD
13.9
Neural Networks and Deep Learning
13.10
Platform.ai: produce high-quality labels
14
ML online books
14.1
Neural Networks and Deep Learning
14.2
Deep Learning
V Examples from Kaggle
15
Examples in Kaggle
16
Melbourne University AES/MathWorks/NIH Seizure Prediction
16.1
Winning solution (1st)
16.1.1
Alex / Gilberto models
16.1.2
Feng models
16.1.3
Andriy models
16.1.4
Code on GitHub
16.2
Solution(4th place)
16.2.1
Pre-processing
16.2.2
Features
16.2.3
Model
16.2.4
GitHub code
17
Bosch Production Line Performance
17.1
1st place solution
17.1.1
Data exploration
17.1.2
Hand crafted features
17.1.3
Hardware
17.2
3rd place solution TBD
17.3
8th place solution with GitHub
17.3.1
Overall architecture
17.3.2
Input data sets
17.3.3
Ensembling
17.3.4
Features
17.3.5
Validation method
17.3.6
Software
17.3.7
Code on GitHub
18
Corporación Favorita Grocery Sales Forecasting
18.1
1st place solution
18.2
4th-Place Solution Overview
18.3
5th Place Solution
19
Rossmann Store Sales
19.1
1st place solution
19.2
3rd place solution
20
Severstal: Steel Defect Detection
21
Lyft 3D Object Detection for Autonomous Vehicles
21.1
3rd place solution
22
APTOS 2019 Blindness Detection
22.1
1st place solution summary
23
Predicting Molecular Properties
23.1
#1 Solution - hybrid
23.1.1
Overall architecture
23.1.2
Ensembling
23.1.3
Hardware
23.1.4
Software
23.1.5
Code on GitHub
23.2
#2 solution 🤖 Quantum Uncertainty 🤖
23.2.1
Overall architecture
23.2.2
Input features and embeddings
23.2.3
Data augmentation
23.2.4
Ensembling
23.2.5
Hardware
23.2.6
Software
23.2.7
Code on GitHub
VI Real world example
24
Real world example
24.1
Subject of the project
Depending from where you were looking:
Looking from the perspective of machine learning expert
24.2
Project phases
The main project phases are:
After data gathering iteration is trump
24.2.1
Feature engineering
24.3
Algorithm selection
24.3.1
Logistic regression
24.3.2
Tree based
24.3.3
Support Vector Machine (SVM) TBD
24.4
Performance measurement
24.4.1
Sensitivity and specificity
24.4.2
Receiver operating characteristic (ROC)
24.5
Confusion matrix and receiver operating characterstic (ROC) for pulse
24.5.1
Receiver operating characterstic (ROC) and probability density plots
24.6
Create augmented labeled data
24.6.1
Features of time signals
24.7
Features generated
24.7.1
Analysis of generated features
24.7.2
Dynamic time warp (DTW) for signal
24.8
Confusion matrix results logistic regression for measured data
24.8.1
ROC results for measured data
24.9
Several algorithms results for SNR = 18dB
24.9.1
ROC results for SNR 18dB
24.10
Calculation of return of invest (ROI)
24.10.1
Calculation of ML project invest
24.10.2
Calculation of ML profit
24.10.3
Resulting ROI
VII Cloud-based machine learning
25
Cloud-based machine learning
VIII Kaggle Survey
26
Kaggle survey introduction
26.1
Kaggle survey details
26.2
Purpose
26.3
Navigation and handling
27
Results
27.1
Survey participants education level
27.2
Who uses which algorithm
27.3
Machine learning experience and algorithms
27.4
Experience and new algorithms
27.5
Role of participants
27.6
Company size
27.7
Company incorporation of machine learning
27.8
Favourite media sources on data science topics
27.9
Favourite online course platform
27.10
Favourite data analyzing tool
27.11
Experience in data analysis coding
27.12
Favourite integrated development environments (IDE’s)
27.13
Favourite hosted notebook products
27.14
Favourite programming languages
27.15
Recommended entry programming language
27.16
Favourite data visualization libraries or tools
27.17
Favourite specialized hardware
27.18
Favourite machine learning frameworks
27.19
Favourite cloud computing platforms
27.20
Favourite big data / analytics products
27.21
Favourite automated machine learning tools (or partial AutoML tools)
References
Published with bookdown
Machine learning orientation
13.3
DeepAI: The front page of A.I.
https://deepai.org
The most popular research, guides, news and more in artificial intelligence