While reading machine learning related papers, I came across these two terms: baseline and benchmark. Although I thought them to be the same thing, there seem to be some differences between them as mentioned here. However, I am still confused in terms of machine learning models, which model to call benchmark and which one to call baseline?
A baseline is usually a lower complexity model. New research leads tend to compare against them to justify increased complexity and other requirements. Often, a random baseline is used, based on either random performance from resampling or a theoretical random performance. This is usually done in applications, not in development of new methods.
A benchmark is a comparison of different competitors. In development, it's often not enough to beat the baseline, the model must be competitive facing other similarly complex models.