Our work demonstrates that approximate Bayesian model selection can be useful in neural nets, even in sophisticated use cases. We aim to further improve the efficiency and usability of neural model selection, making it a more integral part of training deep neural nets. 🧵14/16