Choosing a loss means deciding the price of errors
When deciding which loss function to use for training, we have to understand what the model should consider an expensive error, and what’s a cheap one in the context of the aim of the model.
Rests on
- The model will sacrifice errors the loss considers cheap in order to minimize the errors the loss considers expensive
- A loss function is a preference over error distributions what the loss function is
Cited by
- Log-space target can be used to encode proportional error price across the target distribution an alternative mechanism to target transformation - loss function choice