This project is my humble contribution to the world of deep learning: a neural network that, for unknown reasons, learns in reverse. It starts with great accuracy and gradually forgets everything it once knew.
Essentially, I accidentally invented Machine Unlearning. A true milestone in the history of Artificial Intelligence (or lack thereof).
Here's how well my neural network "learns":
Epoch 1, Accuracy: 86.82%
Epoch 2, Accuracy: 87.07%
...
Epoch 20, Accuracy: 57.97%
Yes, you read that right. Each epoch makes things worse. The poor model started out strong and ended up begging for help. 🤖💔
-
Clone this repository:
git clone https://github.com/your-username/machine-unlearning.git cd machine-unlearning
-
Compile the C code (if you're feeling brave):
gcc -O3 -march=native -ffast-math -o nn nn.c -lm
-
Run it and witness the magic of unlearning:
./nn
-
Sit back and enjoy watching your model's performance decline.
Honestly? Even I’m not quite sure.
Because making mistakes is human, but making a neural network unlearn is a rare talent.
This project is based on the amazing work at miniMNIST-c by Konrad Gajdus. I just accidentally broke it.