acfgm-pytorch

Create a mini-application that demonstrates the power of the 'acfgm-pytorch' package, which implements the Auto-Conditioned Fast Gradient Method (AC-FGM) optimizer for PyTorch. This application will focus on optimizing a simple neural network model using AC-FGM and comparing its performance against standard optimizers like Adam and SGD. Here are the steps and features you need to implement:

1. **Setup Environment**: Ensure your development environment is set up with Python, PyTorch, and acfgm-pytorch installed.
2. **Define Dataset**: Use a publicly available dataset, such as MNIST or CIFAR-10, for training and testing.
3. **Model Definition**: Define a simple Convolutional Neural Network (CNN) model suitable for the chosen dataset.
4. **Optimizer Implementation**: Implement three different optimizers - Adam, SGD, and AC-FGM from the acfgm-pytorch package.
5. **Training Loop**: Write a training loop that trains the CNN model using each of the three optimizers over multiple epochs.
6. **Evaluation**: After training, evaluate the models on a test dataset and compare their performance metrics such as accuracy and loss.
7. **Visualization**: Plot graphs showing the training and validation loss/accuracy for each optimizer.
8. **Discussion**: Include a brief discussion summarizing the observed differences in performance between the optimizers.

This project aims to showcase not only the ease of use of the acfgm-pytorch package but also the potential benefits of using AC-FGM over traditional optimizers in deep learning tasks.