MLP

BP neural network:

1. input layer: input data

2. Hidden layer: responsible for increasing its computing power; how to solve problems. The more layers, the stronger the computing power.

3, output layer: make decisions

Precautions:

1. The more hidden layers there are, the more overfitting occurs. Neurons are connected by synapses.

2. Receive non-numeric data

3. The number of hidden layers is the root number of output layer * input layer.

4. The fewer the number of neurons, the better.

5. Full connection, feedforward, right->left during training

Building a Neural Network

1. Data display

2. Confirm the number of layers and the nodes of each layer

3. Training

4. Results

Bias: Primitive attitude. initial value.

An example:

w/b random give.

Find the partial derivative of the error. Adjust W and b.

Learning rate: Smaller ones learn slower. Big words learn fast.

Neural Networks: Classification, Prediction.

Precautions:

1. The sigmoid function x is between -5 and +5.

2. When BP neural network makes digital prediction, data processing: continuous data must be normalized. Between (-1, 1), the categories should be flattened.

3. BP neural network: perform extreme value normalization for each continuous attribute. or even target properties.

4. Remember to restore the predicted data.

Advantages of Neural Networks:

1. High accuracy

2. When doing test data, the prediction efficiency is high.

shortcoming;

1. The training data takes a long time to learn.

2. Difficulty understanding weights (black box)

3. It is difficult to combine with the field. (purely mathematical model)