Skip to content
Snippets Groups Projects
Commit 9a134aa1 authored by Carel van Niekerk's avatar Carel van Niekerk :computer:
Browse files

Update README.md

parent 8e1da6bc
Branches
No related tags found
No related merge requests found
# Neural Belief Tracking # Code for Paper Knowing What You Know: Calibrating Dialogue Belief StateDistributions via Ensembles
\ No newline at end of file
Implementation of different uncertainty in Deep Learning methods for improved Dialogue ***Belief*** Tracking.
## NBT Model
Slot Utterance Matching Belief Tracker transformer. Using BERT language model fine tuning, a slot utterance matching Attention mechanism and a RNN latent state tracker to generate a latent belief state. Further embedding distance comparison is used for generating model logits.
## Uncertainty
### Loss
* Standard Cross Entropy Loss
* Label Smoothing KL divergence Loss (improved uncertainty)
* Belief Matching (Dirichlet activation function with ELBO Loss)
### Strategies
* Temperature Scaling
* Dropout Ensembling
* Model Ensembling
0% Loading or .
You are about to add 0 people to the discussion. Proceed with caution.
Please register or to comment