StackedML
Practice
Labs
Questions
Models
Pricing
Sign in
Questions
/
Optimization
/
Gradient Methods
/
Convex vs non-convex
← Previous
Next →
704.
Saddle Points in Non-Convex Optimization
medium
Non-convex loss surfaces in deep learning contain saddle points. What characterizes a saddle point?
A
A point where the gradient is large but the loss is low, indicating a steep descent toward a minimum
B
A point where the loss surface is flat in all directions, causing gradient descent to stall indefinitely
C
A point where the gradient is zero but the function curves upward in some directions and downward in others
D
A point where the gradient is zero and the function is at a local minimum along all directions simultaneously
Sign in to verify your answer
← Back to Questions