In addition to my mathematical research, I am very interested in Data Science and Machine Learning, and I hold a Data Science certificate from the Erdős Institute Data Science Boot Camp.
Detecting Images Generated by Neural Networks
I collaborated on a top-5 project (out of approximately 40 teams) to construct two multi-classification models designed to discern between real images and AI-generated ones. Our dual-channel model filtered images using Gaussian blur and Discrete Cosine Transforms before passing them through convolutional and pooling layers. Trained on a dataset of around 90,000 images, our model achieved a precision of roughly 90% and a recall of approximately 93% for detecting real images.
- GitHub: AlvsReal Repository
- Hugging Face: Gradio Interface
- Certificate: Erdős Institute Data Science Boot Camp
Generative Models
To further explore generative AI, I coded a neural network utilizing a variational autoencoder approach to generate digits similar to the MNIST database. This code is based on an implementation of the paper arXiv:1606.05908, utilizing convolutional neural networks which are well-adapted to learning pictures.
- Implementation: Google Colab Notebook
- Reference Paper: arXiv:1606.05908