Quick Check 17.4 – Data 100, Summer 2020

JavaScript isn't enabled in your browser, so this file can't be opened. Enable and reload.

Below are some statements about GD (Gradient Descent) and SGD (Stochastic Gradient Descent). Select ALL correct statements: *

Each iteration of SGD always decreases the loss.

We use SGD for large datasets because it is computationally more efficient per parameter update (runs faster) than GD.

Ignoring issues with numerical precision, gradient descent on the average squared loss is invariant to the sort order of the data.

GD will converge to the same solution regardless of how we set the step size.

None of the Above

Required

(Fall 2018 Final) Which of the following are true regarding GD and SGD? Select all that apply. *

At a particular iteration, stochastic gradient descent will often update θ more accurately compared to an update with regular gradient descent.

For a convex loss function, a single step of gradient descent always decreases the loss.

For a convex loss function, a single step of stochastic gradient descent always decreases the loss.

Suppose it takes t seconds for one update of regular gradient descent. Stochastic gradient descent can usually perform more than one update in t seconds.

None of the Above

Required

Submit

Clear form

This form was created inside of UC Berkeley. Report Abuse

Forms