Features

(Master of Information Science) Pattern Recognition

• The dimension of the feature vector and the distribution of patterns can change depending on the representation method.
  • For example, instead of using a vector of all pixels for an image,
    • Use the color histogram as a feature.
    • Use the shape as a feature.
    • Use the average value of each row as a feature.
  • Various suitable features can be considered depending on the purpose.
• It is necessary to choose a representation that is suitable for recognition.
  • That is what Feature Engineering is all about.
  • Alternatively, automatic compression can be done with methods like Autoencoder.
  • This is called Representation Learning.

---

• In the case of regression models, X_i.

• Even if the impact of a feature on the learning outcome is low, it is not known whether that information is truly worthless.

• The same information may be encoded in another feature that is not being used.

#machine Learning