🧐 This repository is chaotic because it is my first work as a researcher. Sorry for the inconvenience of the unstructured experiment. If you have any questions or something to discuss, feel free to contact me! [email protected]
| Index | Info | Stability of Code |
|---|---|---|
| Experiment1 | simple vae-bart |  |
| Experiment2 | simple beta annealing schedule BART + beta-VAE GRU |
|
| Experiment3 | beta annealing schedule types BART + beta-VAE |
 |
| Experiment4 | Freezing, Linear Probing Issue |  |
| Mini Experiments | Fast try!!! | |
The only successful case is [CNN-DM] Additional Path From Encoder and use it only for the First word with Pretrained
| Model | ROUGE1 | ROUGE2 | RUOGEL | Novel 1 | Novel 12 | Novel 13 |
|---|---|---|---|---|---|---|
| BART | 43.45 | 20.62 | 40.35 | 1.62 | 11.83 | 21.78 |
| Ours | 43.63 | 20.81 | 40.3 | 1.009 | 7.86 | 15.32 |
Check the short paper Encoder_Extract_Transformer.pdf for more information
- 12/19 Implementated Inference bart parallel(spped up!)
- 12/20 Experiment3-Mini2 finished(cyclic annealing)
- 12/21 Implementated Inference bart vae parallel(spped up!)
- 12/22 Close experiment3(There was a problem in the model) and make experiment4
- 12/23 Make Literature Review Category
- 12/24 Gathered all papers
- 12/28 Experiment4 setting
- 2021/01/26 Close the repo