In the presence of local minima, where the derivative is zero, the algorithm might converge to a suboptimal solution instead of finding the global minimum. Additionally, flat regions or plateaus in the loss landscape, where the gradient is close to zero, can also mislead the algorithm. Gradient-Descent-Stuck-at-Local-Minima-18.png

© 2024 All rights reserved

Built with DataHub LogoDataHub Cloud