Hello,
I see you're interested in learning more about the trainNetwork function in MATLAB. This function operates similarly to the process of training a deep learning network. It begins with forward propagation, followed by calculating gradients using backpropagation on the loss function. Subsequently, the weights are updated based on the algorithm you're using, which in your case is 'Adam'. All these steps are automatically handled by the trainNetwork function.
Since Adam is a stochastic optimizer, it processes data in mini-batches. During forward propagation, data is passed in these mini-batches, and all necessary steps are completed before moving on to the next set of mini-batches. However, if you wish to access the weights and biases, you will need to implement a custom training loop in MATLAB and record the weights and biases at each iteration.
You can view the min-batch loss at every iteration and plot the training-loss and validation-loss graph of mini-batches at every iteration using trainNetwork, by setting the plot to "training-progress" in trainingOptions.
Please go through the following documentation to learn more regarding custom training loop and how to implement them,
Please go through the following documentation to learn more regarding 'trainNetwork' function,
Also, currently MATLAB does not recommend 'trainNetwork' function instead you can use 'trainnet' function.
Please go through the following documentation to learn more about how to ask question on MATLAB answer and get a fast response,
Hope this helps.
