From scratch

Implementation of classical deep learning models/algorithms in Python using some common libraries (numpy, torch, etc).

1) FeedForward neural network

Implementation of feedforward neural networks using numpy and tqdm only.
An example is given in the notebook example.ipynb and a mathematical justification is given in the file theory.pdf.

2) Attention is all you need

Implementation of the transform architecture proposed in the paper Attention Is all You Need, encoder and decoder pre-training on French and English datasets, and then fine-tuning for French-to-English translation.

3) Denoising Diffusion Probabilistic Models

Implementation of the diffusion process proposed in DDPM paper and training on the MNIST dataset.
The model used to predict noise (U-Net) is taken directly from Hugging Face's Diffuser library.