https://github.com/shivamsaboo17/GLC
Gold Loss Correction for training neural networks with labels corrupted with severe noise
https://github.com/shivamsaboo17/GLC
classification deep-learning nips-2018 noisy-labels pytorch
Last synced: 7 months ago
JSON representation
Gold Loss Correction for training neural networks with labels corrupted with severe noise
- Host: GitHub
- URL: https://github.com/shivamsaboo17/GLC
- Owner: shivamsaboo17
- Created: 2019-08-17T16:07:59.000Z (about 6 years ago)
- Default Branch: master
- Last Pushed: 2019-08-17T16:24:22.000Z (about 6 years ago)
- Last Synced: 2024-11-05T10:44:36.391Z (12 months ago)
- Topics: classification, deep-learning, nips-2018, noisy-labels, pytorch
- Language: Jupyter Notebook
- Size: 8.79 KB
- Stars: 13
- Watchers: 3
- Forks: 2
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# GLC
Unofficial implementation of [Using Trusted Data to Train Deep Networks on
Labels Corrupted by Severe Noise](https://arxiv.org/pdf/1802.05300.pdf) (NIPS 18) in PyTorch.
## Usage
(See example.ipynb for a walkthrough on MNIST)
```python
from datasets import GoldCorrectionDataset
from glc import CorrectionGenerator, GoldCorrectionLossFunction
c_gen = CorrectionGenerator(simulate=True, dataset=trn_ds, randomization_strength=1.0)
# Fetch both corrupted and clean datasets if in simuate mode
trusted_dataset, untrusted_dataset = c_gen.fetch_datasets()
"""
Train the model on untrusted_dataset
"""
# Generate correction matrix
label_correction_matrix = c_gen.generate_correction_matrix(trainer.model, 32)
# Wrap trusted and untrusted dataset together using GoldCorrectionDataset class
gold_ds = GoldCorrectionDataset(trusted_dataset, untrusted_dataset)
gold_dl = DataLoader(gold_ds, batch_size=32, shuffle=True)
# Modified loss function
gold_loss = GoldCorrectionLossFunction(label_correction_matrix)
"""
Train using gold_ds and gold_loss the model, until convergence
"""
```
## Results
### MNIST
#### Regular training on trusted data (~5% of entire data) -> 61.12 accuracy
#### Gold Loss Correction with 5% trusted -> 95.45 accuracy (All samples in untrusted data (95% of total data) is corrupted by randomly assigning labels)