Uncertainty in LS optimization
Closed this issue · 3 comments
Hi Anastasia, did you take a look at that work from Ping Tan I pointed out on using JtJ as an approximate Hessian?
p.s.
Let's use this issue tracker to keep track of things to do for the paper.
Using it with Edoardo and another student from uvic, works like a charm :)
Thanks a lot!
Could you please not completely rewrite a paper, just proofread it instead? I cannot thank you enough for doing this, but I need to learn writing papers myself at some point. Half of the stuff that I think is extremely relevant is gone! Like the sections settings and parameterization. There is still enough time till the deadline, so we do not have to split the work such that you write the paper. Also, this is not a real paper yes, it is just a draft to show the derivation to Dr. Fitzgibbon.
Above all, could you please return the notation back to original and not change the derivation except if you find a mistake? I huge request from my side. I spent a week full time picking the notation that I am now using. The derivation is not finished anyway. With a different notation and changed equations I will just get confused.
For example I did not use H to denote hessian on line 229 of the original paper because I am already using H for something else. I thought it all through, please return the notation.
The introduction part is simply hidden, not removed! Andrew will want to focus on the part that is rendered right now.
As for notation, I'd suggest you to use macros to call things, because people are used to call hessians with H. So as far as you write \hessian then we can quickly swap to different conventions using different letters later on?
Beside the H is there anything else? Perhaps we can just skype when I arrive at the Uni, in ~30m?