https://github.com/sayakpaul/consistency-training-with-supervision
Contains experimentation notebooks for my Keras Example "Consistency Training with Supervision".
https://github.com/sayakpaul/consistency-training-with-supervision
computer-vision consistency-regularization keras robustness-experiments tensorflow
Last synced: 23 days ago
JSON representation
Contains experimentation notebooks for my Keras Example "Consistency Training with Supervision".
- Host: GitHub
- URL: https://github.com/sayakpaul/consistency-training-with-supervision
- Owner: sayakpaul
- License: apache-2.0
- Created: 2021-03-31T06:36:37.000Z (over 4 years ago)
- Default Branch: main
- Last Pushed: 2021-04-26T01:04:27.000Z (over 4 years ago)
- Last Synced: 2024-12-28T03:39:24.688Z (10 months ago)
- Topics: computer-vision, consistency-regularization, keras, robustness-experiments, tensorflow
- Language: Jupyter Notebook
- Homepage: https://keras.io/examples/vision/consistency_training/
- Size: 1.42 MB
- Stars: 5
- Watchers: 2
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Consistency-Training-with-Supervision
Contains experimentation notebooks for my Keras Example [Consistency Training with Supervision](https://keras.io/examples/vision/consistency_training/). This example also provides a template for performing semi-supervised / weakly supervised learning.
Promising results on [CIFAR-10-C](https://github.com/hendrycks/robustness) with the process shown in the example:
**More things one can incorporate**:
* Incorporate more data during training the student.
* Filter high confidence predictions from teacher during training the student.
* Use recipes like [Stochastic Depth](https://arxiv.org/abs/1603.09382) for training the teacher. The current example uses [Stochastic Weight Averaging](https://arxiv.org/abs/1803.05407) to induce geometric ensembling.
Full-scale experiments are available [here](https://git.io/JO55v).
## Acknowledgements
* [ML-GDE program](https://developers.google.com/programs/experts/) for providing GCP credits.