Bias is a phenomenon that occurs when an algorithm produces results that are systemically prejudiced due to an erroneous assumption.
For example imagine the image on the left is targets being shot with arrows. The target on the lower left has the arrows clumped together (low variance) but a little bit too high to consistently hit the bullseye. It is as if the the wind was consistently blowing all the arrows up. Likewise, bias moves all data points one direction based on its influence. A proper model mitigates the influence of bias.
Human bias can effect data, collection and annotation.
Comments are closed