Adaptive Reweighting of BPINN Loglikelihood

[AstitvaAggarwal]

Checklist

• Tests
• Current API dosent break?
• Documentation
• The new code follows the
  contributor guidelines, in particular the SciML Style Guide and
  COLPRAC.
• Any new documentation only uses public API

Additional context

changes from closed PR #785, solve issue #713

[AstitvaAggarwal]

The PINN reweighing strategy cannot be directly applied to Multi Objective Bayesian Optimization (GradientScale Adaptive weighting works but gives inaccurate results, MinMax weighing uses Flux within and is giving problems on differentiating the Composite Dual type LogPosterior), it will need a completely new implementation. Current Literature on this is also at a early stage. I will still need to understand more to get a working model. cc: @ChrisRackauckas @Vaibhavdixit02

[ChrisRackauckas]

I think some of the Flux issues should be gone because master entirely uses Lux now?

[AstitvaAggarwal]

yes, now that Lux and OptimisationOptimisers is used for MinMax, ill come back to this after #799.