Neural Networks & Gradient Descent

GRADIENT DESCENT
Tune learning rates and visualize error surfaces to help a model minimize loss and converge on the optimal solution.
LOSS FUNCTION1.0000
Step Size (Learning Rate α)