Using machine learning to bridge length- and timescales in atomistic simulations: From OLEDs to chemical reactions