TensorFlow ensembles for multiple-choice-learning.
This research focuses on encouraging useful diversity for Multiple Choice Learning (MCL) problems (see Stefan Lee's NIPS 2016 paper).
We plan to introduce several new methods that represent a paradigm shift for handling MCL, or M-best, problems. Rather than training an ensemble of models using a diversity loss term, our approaches will focus model attention on different parts of the input space.
Structure of Repo:
- Each folder represents a publication.
- Within each folder you will find
- Code
- PDF of final publication
- Jupyter Notebook Tutorial