CS 451 Quiz 24
Recommender systems and collaborative filtering
According to Andrew Ng, recommender systems
receive much attention in academia but are not widely used in industry
receive relatively little attention in academia, but are widely used in industry
Suppose user j has watched movie i and rated it 4 out of 5. Which of the following describes this scenario?
r(i, j) = 0 and y(i, j) = 4
r(i, j) = 1 and y(i, j) = 4
r(i, j) = 4 and y(i, j) = undefined
r(i, j) = 4 and y(i, j) = 5
To learn the movie ratings for user j, we can use
linear regression over all movies i
linear regression over all movies i for which r(i, j) = 1
logistic regression over all movies i
logistic regression over all movies i for which r(i, j) = 1
Collaborative filtering is an algorithm for
One way to implement collaborative filtering is to alternate estimating theta's from x's, and x's from theta's. Which of the following algorithms operates in a similar way?
Instead of alternating the two estimation problems, we can estimate x's and theta's jointly in a single minimization problem.
Why do we initialize x's and theta's to small random values?
This step is optional. Initializing to all 0's would work just as well.
Random initialization is always necessary when using gradient descent
This ensures that x(i) != theta(j) for any i, j
For symmetry breaking, so the algorithm can learn features that are different from each other
In collaborative filtering we don't need to include "bias" features x(0) = 1 and theta(0) = 1
Which of the following is NOT one of movie titles used as examples?
Swords vs. karate
Hot wheels of fire
Cute puppies of love
I'd rather not get a free point on this quiz.
HW 5 teams [not graded]
Please enter one of "working with [your partners name]", "working alone", or "still looking for a partner"
Never submit passwords through Google Forms.
This content is neither created nor endorsed by Google.
Terms of Service