TODO:

• Implement "Attention Is All You Need"
  • Implement encoder and decoder in pytorch for sequence to sequence transduction (as in the original paper)
    • Visualize the self-attention matrices as a heatmap
  • Read the original GPT paper and implement and train a decoder only transformer to understand the basics of how LLMs work.
  • Reimplement the LLM from scratch to fully understand.

Interesting papers:

• Variational autoencoders

• Gaussian splatting

  • 3D Gaussian Splatting for Real-Time Radiance Field Rendering

• CLIP

  • Learn multimodel learning, contrastive learning, embeddings, zero-shot classification.

• SegFormer

  • Learn segmentation and dense prediction.

• Stable diffusion

• Speech to text and text to speech

  • Tacotron
  • Whisper

• AlphaZero