In triple loss are we minimizing the loss function by using gradient descent on the values of A, P, N?
Yes you can use the gradient descent to minimize the loss composed of Euclidean distance.
Right, we are using Gradient Descent as Gent says, but note that what changes is not the values of A, P and N, but rather the parameters of the model (the coefficients that define the function f in the loss formula that Gent has shown). The A, P and N values are just the training data. We don’t actually have to run the training in this assignment, but they show us how to build the cost function so that we understand how FaceNet was actually trained.