Speaker
Description
According to Rich Sutton in his influential essay The Bitter Lesson, "researchers seek to leverage their human knowledge of the domain, but the only thing that matters in the long run is the leveraging of computation." This stark claim runs counter to the instincts of most practicing scientists, for whom domain knowledge and inductive biases have a privileged status. In this talk I will explore how applicable Sutton's Bitter Lesson is to research in computational materials science, drawing on some recent research from our group. I will show how, in certain situations where data is plentiful and covers the domain of application, large models, with loose inductive biases prove (surprisingly) successful. On the other hand, I show how using physics in the process of model design allows us to capture responses of the system that were never exposed during training, hopefully demonstrating that there is still a place for human intelligence in materials modelling, and providing a sweet aftertaste to the bitter lesson.