Doesn’t “baseline” in machine learning mean a very basic model/solution to the problem? and you try to improve on that baseline.
Human level performance(HLP) and state-of-the-art models are too good to be a baseline.
Hi Bassel,
this is more related to the model performance than model structure.
Let’s have medical image classification example where:
- typical human 3 % error
- doctor 1 % error
- experienced doctor 0,7 % error
- team of experinced doctors 0,5 % error
And you decide to take 1% error as baseline and try to beat that value with your model. You don’t have to change the model structure and just by training model on more data you can achieve error less than 1 % = less than your base line.
2 Likes