Predicting duplicate question pairs on Quora using transfer learning.

Performance of recently proposed tranfer learning techniques for Natural Language Processing (NLP) applications (FitLaM and USE) are compared on the Quora Duplicate Question Pair Prediction Challenge. The effect on performance by varying dataset size is also investigated.

Validation Set

1. Negative Log Loss

Model/Training Set Size	100	1000	10000	50000	100000	All
ArCos USE						0.537

Help Needed!

I'm completely packed with work at present and not able to work full time on this project. If you'd like to contribute to this repo, shoot me a mail at rudraksht at iiitd.ac.in