Skip to main content
Theorem
Path
Curriculum
Paths
Labs
Diagnostic
Blog
Search
Sign in
Quiz Hub
/
Score Matching
Score Matching
6 questions
Difficulty 6-8
View topic
Intermediate
0 / 6
1 intermediate
5 advanced
Adapts to your performance
1 / 6
120s
intermediate (6/10)
conceptual
Why does the partition function stop mattering when we train a data-space score field?
Hide and think first
A.
Because the partition function is always equal to one after taking logs
B.
Because score matching replaces the partition function with a sample average over the model distribution
C.
Because the score is a gradient with respect to the input variable, and the partition function does not depend on that variable
D.
Because integration by parts cancels every normalization constant in every likelihood-based objective
Show Hint
Submit Answer
I don't know