Two teams build the same house-price network from the same houses. The only difference is how they record one input, the distance to the city centre:
| model | distance recorded in | a typical value |
|---|---|---|
| A | metres | |
| B | kilometres |
Both models are trained with the same L2 penalty: each weight adds to the loss.
As a thought experiment, suppose both models end up making exactly the same predictions. Then, for every house, the weight on the distance input must produce the same contribution, weight distance, in both models. In model A that weight is .
How does the L2 charge on model B's distance weight compare with the charge on model A's?
Select all that apply.