@ericyocamShipped: GAN-Based Handwritten Digit Generation
An open-source generative AI proof of concept demonstrating how Generative Adversarial Networks (GANs) learn to create synthetic handwritten digit images using the MNIST dataset.
The project implements two competing neural networks: a generator that produces synthetic 28×28 grayscale digit images from random noise, and a discriminator that learns to distinguish generated images from real handwritten digits.
The generator uses transposed convolutional layers, batch normalization, and LeakyReLU activations. The discriminator uses convolutional layers, dropout, and binary classification to evaluate image authenticity.
Both networks are trained through an adversarial learning process, with generated images saved periodically to visualize training progress.
The repository includes an executable notebook and example results, providing a hands-on educational demonstration of generative AI, neural network training, and synthetic image generation.
Designed for AI developers, students, researchers, and educators exploring GAN architectures and generative machine learning.
0 comments