Rigor and Reality

Now the distinction between “theory” and “practice” is often conceptualised as opposing ends of a spectrum. Maybe you consider the former admirable for being pure, maybe the latter for being useful – I mean, people can have a difference in taste here, right, and I’m not interested in arguing one way or the other.

Recently on Twitter I saw Jeremy Cohen make the argument that a one-dimensional view of theory versus practice is not very productive. Theorists aren’t trying to make their work impractical; practitioners aren’t trying to not understand what they’re doing. Ideally, something is theoretically sound and practically useful, so he proposes a 2-dimensional plane with “rigor” and “reality” on the axes, and now the idea is that your research should be on the Pareto front. If your algorithm is less theoretically rigorous, it had better be an improvement in practical reality. If your algorithm doesn’t actually work as well in practice as something else, you’d better have good reasons for doing things that way. (Maybe you can prove something about it that’s not known - or not even true - about the alternatives.)