Adaptive Computation Time: The Remainder
Intermediate
The problem, from first principles
Adaptive Computation Time (Graves, 2016) lets a recurrent network decide how many steps to think before answering. At each step it accumulates a halting probability, and the last step gets the remainder so the weights sum to one.
From theory to code
Implement act_remainder(halts), the weight left for the final step.
Constraints
- The remainder makes the step weights sum to one.
Hints
Hint 1
Subtract the probabilities of all steps except the last from one.
Keywords
research-paperssequence-modelsadaptive-computationrecurrence
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.