Welcome to fusion-opt’s documentation!

Fusion-opt is a framework for applying Meta-learning based bayesian optimization equipped with humnan feedback capability during optimization, to guide the optimization specially in the initial stages and also to have a warm start.

Main contributions of fusion-opt:

• Using state of the art Neural Acquistion process(NAP) as a surrogate for black box functions

• Using preference model trained on experts’ feedback to enable scoring preference of the points to query next

• Suggesting multiple candidate points to human(expert), from which one to be selected as next point to be queried

• A newly designed Acquistion Function combines the preference model and the NAP to suggest one candidate point

• The other candidate points are chosen by desired well-known statistical or Monte-carlo based Acquistion functions such as EI, MES, …

• Modular design, i.e users can utilize pre-defined modules for preference models, or build their own

• Explainability, the framework explains each of the candidate points so that human in the loop can make better decsion for selecting the next point