The dependence of Deep Learning models on randomness is often neglected, which can lead to problems in operational forecasting with extended forecast periods. I am pleased to share our new paper with potential solutions for using this randomness to improve forecast quality. rdcu.be/eX8Xg