Why causal and traditional representation learning need each other

Causal representation learning and traditional representation learning have developed in isolation, creating a terminology gap and missed opportunities. This paper unifies both under a single framework: a task component (what information to preserve) and a constraint component (what structure to impose). Experiments on CausalVerse show causal constraints only help when paired with the right task objective—suggesting neither approach alone is sufficient, but combining them pays off for both theory and practice.