Aug 24, 2023, 3:14:44 AMAug 24
to Ceres Solver
I am currently working on an embedded optimization problem within a Bayesian context and I am employing Ceres Solver for this purpose. Specifically, I'm using a two-loop approach:
1. **Outer Loop**: Minimizing -2 Log Likelihood (-2ll) using numerical differentiation.
2. **Inner Loop (EBE part)**: Computing Empirical Bayes Estimation (EBE) with closed-form gradient.
For the outer loop, the numerical differentiation is functioning well, and the inner loop also successfully computes the EBE.
My query pertains to the inner loop where I have a closed-form gradient. Is it possible to start the direction computation from the approximated Hessian that was calculated at the end of the previous iteration? My objective is to leverage this existing information to enhance the efficiency of the computation.
Any guidance or suggestions on how to implement this within Ceres would be greatly appreciated.
Thank you for your assistance.