←
Back to Gallery
Gradient Descent Landscape
Loss Surface Function
Simple Convex (Quadratic)
Rosenbrock (Banana Valley)
Rastrigin (Many Local Minima)
Beale Function
Saddle Point
Optimizer
SGD
Momentum
RMSprop
Adam
Start
Step
Reset
Learning Rate:
0.01
Momentum (β):
0.9
Step
0
Position (x, y)
(0, 0)
Loss Value
-
Gradient Norm
-
Drag to rotate
the 3D view.
Click on surface
to set starting point.
Watch how different optimizers navigate the loss landscape!
Current Position
Trajectory Path
Global Minimum