Review Note
Last Update: 01/19/2025 05:49 PM
Current Deck: Deep Learning
Published
Fields:
Front
What are the key properties of a good loss function design?
Back
1. It should be differentiable.
2. Derived from the maximum likelihood principle to simplify optimization.
3. Should consider the output of the neural network as parameters of a distribution over the targets.
2. Derived from the maximum likelihood principle to simplify optimization.
3. Should consider the output of the neural network as parameters of a distribution over the targets.
Suggested Changes:
Deck Changes (Suggestion to move the Note to the following Deck):
Field Changes:
Tag Changes: