GoogLeNet: Auxiliary Classifier Loss
Intermediate
The problem, from first principles
GoogLeNet attaches two auxiliary classifiers to middle layers and adds their losses to the main loss with weight 0.3. The extra signal helps gradients reach the early layers of a very deep network.
From theory to code
Implement weighted_total_loss(main, aux, w), returning the main loss plus the weighted auxiliary loss.
Constraints
- The default weight is 0.3, as in the paper.
Hints
Hint 1
Multiply the auxiliary loss by w and add it to the main loss.
Keywords
research-paperscomputer-visioncnninception
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.