What does the "no free lunch" theorem say about learning algorithms?
- No model is best on every problem
- Bigger models always generalize better
- Every model needs labeled data
- Training cost grows with accuracy
Averaged over all possible problems, every algorithm performs equally, so model choice must fit the data at hand.
Play AI & ML ▸