Are there any multi-objective optimization capabilities in auto-sklearn?

I use AutoML, and most of the time, I am not interested in differences of 1-2% in a performance metric, but care about things like: Minimizing the number of input features, building simple models that are highly interpretable, or building models with a fast inference speed.

My question is - are there any ongoing efforts to run a model search optimizing not just for performances, but for simple (e.g. low feature, low parameter) models? or fair models? or fast inference speed models? If not, I wonder if anybody building auto-sklearn would be interested in discussing the concept.

-Will David, wdavid2222@gmail.com