How do modern optimizers reshape cost function geometry?

Optimizers like Adam and SGD implicitly transform the effective geometry of the loss landscape.

Optimizers do not just follow gradients—they modify the effective cost surface. Adam rescales gradients using adaptive moments, while SGD introduces noise that biases convergence toward flatter minima. This means the optimizer defines an implicit geometry over the original cost function.