This repository attempts to implement in code a model that learns to forget. By implementing a simple text classifier model on a wine-review dataset, randomizing the weights of a trained model can be used as a method to make a trained model forget what it has learnt.
| Layer (type:depth-idx) | Param |
|---|---|
| TextClassifier | -- |
| └─Linear: 1-1 | 356,160 |
| └─Linear: 1-2 | 19,110 |
| └─ReLU: 1-3 | -- |
| └─Dropout: 1-4 | -- |
Total params: 375,270
Trainable params: 375,270
Non-trainable params: 0
Original Model
| Loss | 0.6329621076583862 |
| Train Accuracy | 0.8930568099021912 |
| Validation Loss | 0.6247082948684692 |
| Validation Accuracy | 0.9080535769462585 |
Randomized Weights Model i.e the model that forgot
| Loss | 3.1483829021453857 |
| Train Accuracy | 0.507146954536438 |
| Validation Loss | 3.06455397605896 |
| Validation Accuracy | 0.508832573890686 |
Comparison Graph
