Learning by association; Semi-supervised MNIST learning
This is an implementation of "Learning by Assocication" written with Keras.
assoc.py execs semi-supervised learning of MNIST.
usage
Type make help.
Result
| #labels | Acc.(%) |
|---|---|
| 100 | 96.00 |
| 10 | 90.30 |
The paper reported accuracy 99%+ can be performed with 100 labels (https://github.com/haeusser/learning_by_association/blob/master/semisup/mnist_train_eval.py). This project of mine couldn't achieved the accuracy...
References
- The original paper: [1706.00909] Learning by Association - A versatile semi-supervised training method for neural networks
- Source code by them: https://github.com/haeusser/learning_by_association
- 日本語の解説: arXivTimes/arXivTimes#352