diff --git a/README.md b/README.md
index 7351205..5a09b49 100644
--- a/README.md
+++ b/README.md
@@ -53,10 +53,13 @@ python3 src/run.py train.pl_trainer.gpus=1
### Omniglot (5-way 5-shot)
+Few-shot learning using this dataset is easy task to overfit or learn for
+MAML algorithm.
+
- |
+ |
Metatrain |
Metavalidation |
@@ -75,19 +78,19 @@ python3 src/run.py train.pl_trainer.gpus=1
MAML |
OmniConv |
1 |
- |
- |
- |
- |
+ 0.992 |
+ 0.992 |
+ 0.98 |
+ 0.98 |
MAML |
OmniConv |
5 |
- |
- |
- |
- |
+ 1.0 |
+ 1.0 |
+ 1.0 |
+ 1.0 |
@@ -157,6 +160,13 @@ Run the experiment as follows:
python3 src/run.py data=miniimagenet
```
+
+## Implementing a different meta learning algorithm
+
+If you plant to implement a new variant of MAML algorithm (for example
+MAML++) you can start by extending [default lightning module](https://github.com/rcmalli/lightning-maml/blob/44f271380bb6efc925a9070abe2ec4d0f7d88ef3/src/pl/model.py#L77) and its [step](https://github.com/rcmalli/lightning-maml/blob/44f271380bb6efc925a9070abe2ec4d0f7d88ef3/src/pl/model.py#L100-L150)
+function.
+
## Notes
There are few required modifications run meta-learning algorithm using