Fixed point iterations for parallelizing nonlinear dynamics is all the rage: - Newton for RNNs - Picard for diffusion models - Jacobi for parallel decode of LLMs
But how do these techniques relate, and when should you use them?
We show you how in our new paper 🧵



