shuida/build_deep_learning_framework

a simple implement for building deep learning framework

Build Your deep learning framework

Construct your own deep learning framework,which help you to be familiar with forward propagation, backward propagation, and update model parameters by gradient descent method. This is a very simple deep learning framework, which is conducive to understand model and optimization.

Function

• SGD_Linear_regression: We use Least-Squares method and the SGD optimization model to perform linear regression,and compare the difference between them.
• LearnConv_TraditionCV: We use the traditional method to transform the color of the image. At the same time, we use the network to learn the color transformation, and compare the difference between the two.

Code Structure

Conv and Fc

For every layer, we need to code the forward function and the backward function.

• the formula doesn't errors
• the relationship between dimensions is not wrong(Do not use reshape at will).

optim

Method of updating parameters based on weight and gradient.

model

This class has the following main effects:

• Contact forward and backward(cache)
• store the parameters of model

solver

This class has the following main effects:

• Save the parameters needed for optimization, such as the momentum of each variable, the order of magnitude of each variable.