Neural networks (neural networks)
1, the background to the issue:
In computer vision processing, picture 50 * 50 has 2500 pixels, for all, "the X- i * the X- J has 2500 * 2500/2 ≈ 300 Wan quadratic term eigenvalues", features lead to excessive over-fitting.
2, model:
If the first layer j S j neurons, the first layer S j + 1 j + 1 neuron, then [theta] (j) is S j + 1 * (S j + 1'd) dimension.
3, mathematical definition:
Positive incentive formula:
4, for example:
5, multi-class classification:
When H [theta] (X) = [. 1 0 0 0] T , for pedestrians;
When H [theta] (X) = [0. 1 0 0] T , for the car;
When H [theta] (X) = [0 0 0. 1] T , for the motor;
and many more
The end result of the form:
