Lorenzo Stella

I think it’s fine to have no AD-backend by default (which would make the package more lightweight on dependencies as @gdalle proposes). In this case, the [fallback definition for gradients](https://github.com/JuliaFirstOrder/ProximalAlgorithms.jl/blob/3de04a9b2878925f8307f756d66fbe528afb2e5a/src/utilities/ad.jl#L4)...

@Maximilian-Stefan-Ernst I'm not sure how Optim deals with `Inf` values. From what you say, I'm guessing you are trying to deal with constraints in some smooth term (having it return...

> Sorry for the long post, but I could not find an easier MWE that reproduces the problem. Not at all, thanks for the detailed report! Leaving aside the AD...

The way I understand the current AD ecosystem (without looking much into the packages though): * ChainRulesCore defines the API and core utils for differentiation rules. * ChainRules defines differentiation...

@mohamed82008 thanks, it seemed to much of a general issue not to have a solution already :-) now I’m curious, I’ll look into that

@mohamed82008 ProximalAlgorithms supports nonconvex problems, in the sense that some of the implemented algorithms don’t need convexity assumptions on the objective terms in order to converge. AD (Zygote right now)...

cc @nantonel @mfalt

@fabian-sp I think it would be useful to have them in the package. There could be multiple solutions for indexing the groups: I guess the simplest (and most general?) thing...

Well if they’re needed then they will be checked out sooner, otherwise they will stay longer on the list :-) Added, thanks

Thank you @baggepinnen, I’m assuming this is in the context of [IndBallRank](https://kul-forbes.github.io/ProximalOperators.jl/latest/functions/#ProximalOperators.IndBallRank) or [NuclearNorm](https://kul-forbes.github.io/ProximalOperators.jl/latest/functions/#ProximalOperators.NuclearNorm) or both?