Baseline level of performance

Hi everyone, I’m doing currently the third week of Advanced Learning Algorithms, and can’t understand two things about baseline level of performance. The first one is: if we want to use human level performance, does it matter which data set(training or cross validation set) to use to calculate this baseline level? And if it’s just taking performance of some other algorithm, should we take training error or cross validation error of that algorithm as a baseline level of performance for our algorithm?

And the second thing is: why are there barely any search results related to baseline level of performance?

Link to video:

Thanks in advance:)

It’s not a popular concept, because it’s impossible to quantify and very expensive to measure.

1 Like

Ok, thank you :slight_smile: