Preprint of my first PhD paper!
Published:
The first paper that I worked on during my PhD is now available on HAL and arXiv. It includes an error model for Newton’s methods, when inexactness can arise due to different types of approximations, and the derived convergence theory. What both theory and experiments tell us is that we can use low precision formats to solve Newton’s system almost for free in the majority of cases. If you work on 2nd order optimization, or (wisely) plan to do so, check it out!
