I have a sort of “philosophical” question about model training. Which model will be more accurate and useful, generalist or specialist? My intuition says specialist, but I’d like to hear to the community thinks.
For example, I work with audio data, and have recordings from one specific location where I’d like to detect a signal of interest. So I train my model on data from only this location and nowhere else. This “specialist” model should perform well for this location.
But what if I use data from a number of locations, technically, the model will have more variations to learn, but is it really better for the accuracy?