arxivcs.AI2026-07-06
MoP-JEPA: Hard-Assigned Predictor Mixtures for Stochastic JEPA World Models
Zhi Song, Ximing Xing, Zhenchao Tang, hanbo Huang, Weilong Yan, Tianxu Lv, et al.
JEPA world models commonly predict the next latent state with one regressor. Under stochastic transitions, squared and cosine regression return the conditional mean and its normalized direction, respectively: a single compromise that may match no valid successor. MoP-JEPA instead…