Vokenization Explained!

Vokenization Explained!

Nov 10, 2020
|
41 views
|
|
Code
Details
To solve common sense issues, new approaches have been proposed by University of North Carolina, Chapel Hill researchers to visually supervise language models to achieve performance gains, called Vokenization - Only for GLUE benchmark and SQuAD. This approach is done by creating token-image matching (vokens) and then classifying corresponding tokens with a weakly-supervised loss function.

Comments
loading...