1
$\begingroup$

I am a copyright scholar so please forgive my ignorance. When weights are stored external to a model what is the mechanism by which the weight knows which neuron or node in a decision tree it is attached to when ingested back into the model?

$\endgroup$
0

1 Answer 1

0
$\begingroup$

Neural networks are not composed of "neurons" or "decision trees". Neural networks are a bunch of matrix operations, e.g. matrix multiplications. Some of these operations have trainable parameters, e.g. in a matrix multiplication, the matrix by which you multiply the input is trainable. When a model is saved to file, both the specification of which operations are applied (and in what order they are applied) and the trainable parameters are saved (e.g. multiplication by matrix with reference "weight_f1j390" + "weight_f1j390=[[0, 3, 1],[1, -1.2, 0.5]]).

When the model is loaded back, the sequence of operations is reconstructed from the file together with the values of the weights.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.