Unofficial reproduction of the paper What's Hidden in a Randomly Weighted Neural Network?.
Status: in development.
Dataset | Arch | Epochs | edge-popup | Commit | Accuracy | # Runs |
---|---|---|---|---|---|---|
Imagenette | xresnet50 |
5 | yes | 55e34a6b | 76.18% | 5, mean |
Imagenette | xresnet50 |
5 | no | 55e34a6b | 80.94% | 5, mean |
Imagenette | xresnet50 |
20 | yes | 55e34a6b | 86.44% | 5, mean |
Imagenette | xresnet50 |
20 | no | 55e34a6b | 89.53% | 5, mean |
Paper (CIFAR-10 , Conv 8 ) |
Reproduction (Imagewoof , xresnet50 , 2f1f0b0d) |
---|---|
- init
- signed kaiming constant
- kaiming normal
- layers
-
Linear
-
Conv2d
-
BatchNorm2d
-
LSTM
-
- tests
- initializations / variance
Any contributions are welcome.
Feel free to file an issue
or send a pull request
.