How to reduce the dimension of the task for training a neural network?

Question

Neural networks

The input is a 70x80 matrix filled with 0's and 1's. The output is a vector of 6 values ​​showing which of the 6 given classes the matrix most likely belongs to. When using a line-by-line sum or a sum by columns, it turns out that there should be 76 (86) neurons on the input layer of the NN. How can the dimension of the problem be reduced?

Answer 1

You can try something standard like PCA. But it would be better to get information about the nature of the possible filling of the matrix and come up with your own features. Line sum is one option.
An input layer of 5000 neurons does not mean that the dimension should be reduced. It must be reduced if, due to the length of the vector, training takes an unacceptably long time.