Memory Networks: Matching the Question to Memory
Beginner
The problem, from first principles
Memory Networks (Weston et al., 2014) store sentences as memory embeddings. To answer a question, the model scores every memory against the question embedding, then reads the memories that match.
From theory to code
Implement memory_match(u, M), the dot product of the question with each memory row.
Constraints
- One score per memory row.
Hints
Hint 1
Multiply the memory matrix by the question vector.
Keywords
research-paperssequence-modelsmemoryquestion-answering
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.