1 The Perceptron Model
it just a linear model where:
Here \(\phi(x)=x\) for instance.
1.1 Error function
1.2 The Perceptron: Learning
Here \(<1\) we can also have \(\gamma\)
1.3 Perceptron Learning as Gradient Descent
1.4 Pros with the Perceptron
- The algorithm guarantees to converge if the data is linear separable
1.5 Problems with the Perceptron
- Perceptron only works for 2 classes
- Cycling theorem: many solutions if data is not linearly separable
- Based on linear combination of fixed basis functions.