Facundo Sapienza

This thread is to discuss how to introduce symbolic differentiation in the review paper. It may be worth noticing that symbolic differentiation actually gives a different result that AD, specially...

Existing software to compute sensitivities

2

This thread is to add software that compute sensitivities though ODE solver at least with one of the methods explained in the review paper. - `CVODES` in Sundials - `FATODE`...

There is a reference by Lev Semenovich Pontryagin in "mathematical theory of optimal processes" (1967) with derivation of the adjoints. It is an old reference, but it has all the...

I am creating this issue to add documentation that treats the use of DP tools in geophysical models, reviews and concrete applications. - [Differentiable modeling to unify machine learning and...

Make more explicit the comparison between backprop in NN and adjoints introduced in the original NeuralODE paper. Also emphatize that there are naturally not the same!

Currently, every-time someone does a PR to main there is a conflict in the `main.pdf` file that prevents the auto-merge and it cannot be solved directly from GitHub, but instead...

This paper includes a short comparison between discrete and continuous adjoints that it may be worth referencing in the review.

The adjoint can be approximated using minimax arguments on the lagrangian. Check if this could be useful at the moment of computing fair approximations of the Jacobian.

Notice that in the discrete adjoint method, we have the term $$\frac{\partial A}{\partial \theta}$$ For $A$ matrix and $\theta$ vector this results in a tensor that may be difficult to...

This is already there, so we don't need an extra section about it if instead, we make clear that the current formulation for adjoints actually includes the PDE case.