Machine Learning

Table of Contents

Introducation

Machine learning (ML) is a subfield of artificial intelligence (AI).

Application

Image processing

  • Image tagging
  • Optical character recognition (OCR)
  • Self-driving cars

Text analysis

  • Spam filtering
  • Sentiment analysis
  • Information extract

Data mining

Discover patterns and make predictions from data.

  • Anomaly detection
  • Association rules
  • Grouping
  • Predictions

Video games & robotics

Algorithms

  • ML algorithms try to solve the same problem: learn to transform every input to the correct output.
  • ML algorithms try to make generalizations, and create the simplest hypothesis (with fewest assumptions) that explains most of the training examples.

Categories

Supervised learning

For each training example, the correct output must be provided.

Unsupervised learning

The training examples but not the desired output are the input.

Algorithms

Support vector machines (SVM's)

The model tries to build a set of hyperplanes in a high dimensional space which tries to separate instances of different classes by getting the largest separation between the nearest instances from different classes.

Probabilistic models

  • The models try to predict the correct response by modeling the problem with a probability distribution, e.g., Naive Bayes classifiers.
  • It returns the prediction as well as the corresponding degree of certainty.

Deep learning

  • It is based on the artificial neural network models. They try to emulate (in a very simplified way) the way the brain works.
  • The network consist of a huge set of interconnected neurons, i.e., the basic processing unit, organized in various layers.

Feature engineering

Feature extraction

Before fed to the ML algorithms, the raw data should be tranformed into a vector of features.

Feature selection

  • Score each feature and return only the most valuable ones.
  • Do not use huge feature sets.

Performance metric

  • The instances for training and testing must be disjoint.

Deep learning

A subfield of machine learning.

Recurrent neural network (RNN)

Convolutional neural network (CNN)

  • Take a 2-D array as input.
  • Direct operation on the image rather than feature extraction
  • Basic ideas
    • Local receptive field
    • Convolution
    • Pooling