Deep Learning - Deep Neural Network for Beginners Using Python - Local Minima Problem

The video tutorial discusses the concept of local and global minima in the context of error functions and gradient descent. It explains how error functions can have multiple local minima, which can mislead gradient descent algorithms into thinking they have found the point of minimum error. The tutorial highlights the challenge of reaching the global minima, where the error is truly minimized, and discusses the implications of this in training models.

What is the main difference between a simple ball-shaped error function and a complex one?

• A simple ball-shaped error function has multiple minima.

• A complex error function has only one minimum.

• A simple ball-shaped error function has only one minimum.

• A complex error function has no minima.

In the context of error functions, what is a local minimum?

• The deepest point in the error function.

• The point where the error is maximum.

• A point where the error is zero.

• A point where the error is lower than its immediate surroundings.

How is a global minimum different from a local minimum?

• A local minimum is the deepest point in the error function.

• A local minimum is the highest point in the error function.

• A global minimum is the deepest point in the error function.

• A global minimum is the highest point in the error function.

What challenge does gradient descent face when identifying minima?

• It can mistakenly identify a local minimum as the global minimum.

• It always finds the global minimum.

• It never finds any minimum.

• It only works with simple error functions.

Why might an algorithm incorrectly assume it has found the point of minimum error?

• Because the error function is linear.

• Because the error is maximum.

• Because the gradient is non-zero.

• Because the gradient is zero at a local minimum.