Short Definition
An optimization landscape is the shape of the objective function over possible parameter values.
Intuition
It is the terrain that training moves through: valleys, slopes, flat regions, and rough patches.
Technical Definition
The landscape describes how loss or reward varies across parameter space, including local minima, saddle points, curvature, and flat regions.
Example
Gradient descent moves through the loss landscape while updating neural network weights.
Common Misunderstandings
Deep learning landscapes are high-dimensional and cannot be understood from simple 2D plots alone.
A local minimum is not always a practical problem in modern neural networks.