Schema-Adaptive Action-Conditioned JEPA for Cross-Machine CNC Transfer under Partial Sensor Overlap

A schema-adaptive action-conditioned Joint-Embedding Predictive Architecture (SAAC-JEPA) for CNC dynamics achieved zero-shot RMSE of 0.546, R² of 0.012, and NLL of 0.52 on a sealed target machine sharing only 10 of the source's 17 sensor channels, according to an arXiv paper (2609.16071v1). Across five seeds, JEPA pretraining produced no clean-source forecasting gain, with scratch and pretrained-body models at RMSE 0.811±0.022 and 0.813±0.022, and the locked model underperformed RevIN-equipped PatchTST and iTransformer baselines at 0.503 and 0.498. A pre-declared paired ablation showed RevIN in the same architecture improved RMSE to 0.495±0.004 over three seeds but degraded target calibration to NLL 20.6 on stationary context windows, leading the authors to conclude that source-domain forecasting accuracy alone is insufficient to assess industrial predictive representations.

arXiv:2609.16071v1 Announce Type: new Abstract: Cross-machine deployment of industrial world models requires transfer across changes in dynamics, sensing interfaces, sampling regimes, and control units. We study a schema-adaptive action-conditioned Joint-Embedding Predictive Architecture SAAC-JEPA for CNC dynamics, where the source machine has 17 canonical sensor channels and the target shares only 10. Evaluation uses group-disjoint source splits, source-only normalization, held-out self-supervised validation, unit audits, and a sealed target test after model locking. Across five seeds, JEPA pretraining gives no clean-source forecasting gain: scratch and pretrained-body models obtain $\mathrm{RMSE}=0.811\pm0.022$ and $0.813\pm0.022$. A source-only search over 20 candidates selects a schema-consistent action-conditioned JEPA after seven-seed stability checks. On the confirmatory target pass, the locked model reaches zero-shot $\mathrm{RMSE}=0.546$, $R^2=0.012$, and $\mathrm{NLL}=0.52$, outperforming persistence but not RevIN-equipped PatchTST and iTransformer baselines $0.503$ and $0.498$ . A pre-declared paired ablation shows that RevIN in the same architecture improves RMSE to $0.495\pm0.004$ over three seeds, but degrades target calibration $\mathrm{NLL}=20.6$ on stationary context windows. A pre-lock adaptation sweep further reduces RMSE to $0.520$ with limited target support. These results show that source-domain forecasting accuracy alone is insufficient to assess industrial predictive representations, and that cross-machine adaptation under partial sensor overlap is a distinct evaluation axis.