Machine Learning

Explained to your Grandma

Quentin Fayet

What ?

"Field of study that gives computers the ability to learn without being explicitly programmed"

- Arthur Samuel

What ?

Based of mathematical fields of study :

 

  • Probabilistics
  • Statistics
  • Decision theory
  • Information theory
  • Control theory
  • Optimization
  • Connectionism
  • ...

Why ?

  • Finance : Automated trading, fraud detection
  • Insurance : Fraud detection,
  • Artificial Intelligence
  • Security : Facial recognition
  • Aeronautics : Incident prevention

Resolving problems where human brain is not enough efficient.

How ?

Inputs

Trained model

Outputs

Model

Mathematical law / abstraction of a problem

Training

  • Supervised : Each set of inputs comes with the expected outputs
  • Unsupervised : Expected outputs are not provided
  • Semi-supervised : Some inputs have expected outputs, some don't

Over / Under fitting

  • Overfitting : The model is too specific to the training data
  • Underfitting : The model is too generalistic

Features

Measurable property contained into an input

What kind ?

Machine Learning algorithms belong to different families :

  • Regressions
  • Clusterisations / classifications (binary/multiclass)
  • Regularizations
  • Decision trees
  • Dimensionality reduction
  • Features selection
  • Neural networks
  • ...

Machine / Deep Learning

Machine Learning

Deep Learning

Deep Learning

Attempt to provide high-level abstractions of the data

Dataviz

One of the most important thing to do before using machine learning is to visualize the data.

Dataviz(ualization) is almost a science itself.

Visualization of 2D/3D data is easy. What about 100D data ?

Dataviz

What tools ?

Specialized languages :

  • R (language for statistical processing)
  • Matlab / Octave (Language for mathematic processing)

 

Languages and libraries :

  • Python :
    • Matplotlib (plotting)
    • Numpy / SciPy (science computing)
    • Scikit-learn (Machine / Deep learning)
    • Pandas (Data analysis)

Resources

  • Andrew Ng : Machine Learning (Stanford University)

https://coursera.org/learn/machine-learning

 

  • Machine Learning Summer School

https://www.youtube.com/playlist?list=PLZSO_6-bSqHQCIYxE3ycGLXHMjK3XV7Iz

 

  • KDnuggets

http://www.kdnuggets.com/

 

  • LyonTech Hub (Slack)

Resources

Available at my desk :

  • Pattern Recognition and Machine Learning (Bishop)
  • The Elements of Statistical Learning (Trevor Hastie, Robert Tibshirani, Jerome Friedman)
  • Data Science from Scratch 1Ed (Joel Grus)
  • Machine Learning for Hackers 1Ed (Drew Conway, John Myles White)