Capstone Week 9 Learning Journal (Week Starting February 25)

 This week was continuing the fast ai part 2 course and learning much more. This half is where things got more serious and I learned how to train models faster and more reliably (accelerated SGD tricks, ResNets, dropout, mixed precision), and then I went deep on diffusion and learned better sampling methods with ideas from Karras. I also learned practical applications like super-resolution. To wrap it up, I studied attention/transformers and how all these pieces come together in latent diffusion which is the efficient trick that makes modern text-to-image systems possible.