MLP (Multi-Layer Perceptron)

Training of MLP

MLP learns weights and biases to minimize the difference between actual response and neural network predictions.

$${\mathrm{H}}_k^i=\sum _{j=1}^n{w}_j\cdot {\mathrm{H}}_{k-1}^j+\theta \cdot f_{act.}({\mathrm{H}}_k^i)$$

$w_1,w_2,\cdots ,w_n$ : weight, θ : bias, $f_{act.}$ : activation function

Advantages

- Expresses the nonlinear relationship of input and output variables well.
- Able to create a meta model regardless of a lot of data.

Disdvantages

- User know-how dependent variables → the number of hidden layer / neuron, the activation function form.
- Time consuming for training to create.

Source : https://www.pidotech.com