Yihping Luh

What Persists Matters More Than What Changes

A reusable world model need not reproduce every future observation exactly. What matters is which distinctions remain stable when forecasts, tasks, or trained instances change. Independently trained models may give different values yet preserve the same relation among matched int...

Nonantipodal Semantic Change as a Planning Abstraction for World Model Latents

Planning with a world model depends not only on predictive accuracy but also on which distinctions remain available to downstream reasoning. The abstraction asks whether opposing semantic changes should be merged into one signed coordinate or preserved as separate directions unti...

Withdrawal Constrained Transfer for Revising Rollout Support in World Model Planning

Predictive world models organize candidate futures, yet execution experience can change which futures should retain deployment authority. This paper studies revision of a rollout admission rule rather than construction of a new planner or safety objective. Withdrawal Constrained...

Compositional Reuse of Semantic Change Relations in World Model Dynamics

Compositional generalization asks whether familiar structure can still be reused when familiar pieces appear in combinations absent during fitting. In world model dynamics, this becomes a question of whether directed effects among semantic changes can be recombined without learni...