Neural networks need to learn:
From biological neurons to artificial neurons
Activate functions Relu, Tanh, Sigmoid.
Understanding logistic regression classification through neural network topology
Understanding Softmax regression classification through neural network topology
Solving the problem of upgrading and dimensionality reduction through neural network hidden stratification
The reason why the activation function of hidden layer must be nonlinear is analyzed
Application of neural network in sklearn module
Case study on cement strength prediction and drawing neural network topology
BP back propagation algorithm needs to learn:
BP reverse propagation purpose
Chain derivative rule
BP back propagation derivation
Application of Different Activation Functions in Back Propagation
Application of Different Loss Functions in Back Propagation
Python realizes the actual combat case of neural network
TensorFlow deep learning tools are designed to:
TF installation (including CUDA and cudnn installation)
TF realized the analytical solution of multiple linear regression.
TF realizes the gradient descent solution of multiple linear regression
TF forecast California housing price case
TF realizes Softmax regression.
Case study of MNIST handwritten numeral recognition project
Save and load TF framework model
8) TF has realized DNN multilayer neural network.
9) DNN classification MNIST handwritten numeral recognition project case
10) tensor board module visualization
These are some knowledge involved in deep learning. Generally speaking, it is necessary to deeply understand the neural network algorithm and its optimization algorithm, master the TensorFlow development process, and complete the regression and classification tasks by implementing neural networks. TensorFlow framework is easy to learn, and other deep learning frameworks such as Keras and PyTorch are easy to master. In addition, you can do some actual combat, so that you can be more skilled.