What's the intuition on using "bias" as a description of a model

I understand the “variance” name as it is the divergence (i.e variance) between the results from training data and test data. If it has high variance, then the model is overfitting on training data and not generalizing well.

But what is the deal with “bias” naming? Besides it being what the community has agreed on, I’d also like to get a better intuition as that makes it easier to remember.

The reason is lost to history.

You might consider “bias” as meaning “tends to give the same answer regardless of the inputs”.

Ha! Well I can get on board with your reasoning :smiley:

That said, I think the naming is unfortunate due to the different connotations that word can imply in general use, let alone the multiple uses in ML itself … oh well, guess it’s just something that will stay :man_shrugging:

Hello Maxim, I like how this graph represents variance and bias, how do you think? I think we can consider a dot as a model.