Article

Causal world models under physical parameter shifts: explicit affine coefficients improve extrapolative prediction

This study examines the degradation of world-model predictions when physical conditions fall outside the training distribution, comparing two causal modeling elements: the inputs available to the predictor and how its mechanism computes their effects. On a synthetic benchmark with shortcut features, a frozen dependency mask derived from paired interventions reduced the out-of-distribution RMSE of a fully connected predictor from 1.769 to 0.697, close to the generator-oracle mask's 0.694.

Part of the event: Research on Causal World Models under Physical Parameter Shifts

This page provides a summary and source link only.