Skip to content

Latest commit

 

History

History
21 lines (17 loc) · 663 Bytes

File metadata and controls

21 lines (17 loc) · 663 Bytes

Perceptrons

Neural networks:

  • Artificial neural networks.
  • They use all the points to make a decision.

Perceptrons:

  • 1 → w0, x1 → w1, ..., xn → wn... then go to o = 1, if w0 + sum wi*xi > 0; 0 otherwise.
  • Binary classification.

Perceptron training rule:

  • Randomly initialize weights.
  • Interate through trainint isntances until convergence.
  • Update each weight: Delta wi = etha (y-o)xi.
  • y is the label, ŷ is the predicted label, o is output, η is the learning rate.
  • wi <- wi + Delta wi.

Representional power of perceptrons:

  • Perceptrons can represent only linearly separable concepts.
  • There is a decision boundary.
  • Also as xw > 0.