Neural Turing Machines: Erase and Add Write
Intermediate
The problem, from first principles
The NTM writes by erasing part of each addressed row and then adding new content, both weighted by the write distribution. Erase and add are separate so the model can overwrite a slot cleanly.
From theory to code
Implement ntm_write(memory, w, erase, add), the erase-then-add update.
Constraints
erasevalues are between 0 and 1.
Hints
Hint 1
Scale each row's old content by one minus w times erase, then add w times add.
Keywords
research-paperssequence-modelsmemoryneural-computation
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.