This repo contains the sample implementation for this article:
https://medium.com/p/c9528849760a
https://github.com/lorenzob/curiosity/blob/master/docs/README.md (GitHub version)
The basic idea is to give a higher priorty to the harder samples in the data set, by training on these more often. See the article for details.
NOTE: I'm not yet sure if I did a big silly mistake somewhere, I double checked everything a few times but...you know. Please let me know.
Code
Samples are provided for tensorflow and keras (eager mode) and pytorch.
Current examples:
- MINIST
- Fashion-MNIST
- CIFAR-10
- Linear regression
There is also one implementation of the "pool" idea (mnist_eager_pool.py).
This is the code used to make the sample charts in the article.
Please let me know what do you think, if it works, if there are mistakes, suggestions, etc.