Show, Attend and Tell: Doubly Stochastic Penalty
Advanced
The problem, from first principles
Show, Attend and Tell adds a regularizer that encourages the attention to cover the whole image over the caption. Summed over the words, each location's attention should be about one, which is a doubly stochastic constraint.
From theory to code
Implement doubly_stochastic_penalty(alpha), the squared deviation of each location's total attention from one.
Constraints
- Columns of alpha are locations.
Hints
Hint 1
Sum the attention matrix over the words for each location, subtract from one, square, and sum.
Keywords
research-paperssequence-modelsattentioncaptioning
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.