snowkylin / ntm

TensorFlow implementation of Neural Turing Machines (NTM), with its application on one-shot learning (MANN)

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

What is key vector?

albertchristianto opened this issue · comments

Hi,

This is a great work. I have read your ppt and it is very clear. I also have read the paper. But I have a question. What is kt (key vector)? How to get key vector? because the paper doesn't mention it clearly.

Best regards,
Albert

For content-based addressing, key vector is the "content" that is compared to each vector M_t(i) by a similarity measure. See section 3.3.1 of the NTM paper for details. It is one of the parameters of head and can be generated simply from the output of controller by some linear transformation. See line 29-47 and line 63 in ntm_cell.py for details.

Okay, thank you very much for the response and the information @snowkylin . I will check it.
best regards,
Albert