{"slug": "schema-adaptive-action-conditioned-jepa-for-cross-machine-cnc-transfer-under", "title": "Schema-Adaptive Action-Conditioned JEPA for Cross-Machine CNC Transfer under Partial Sensor Overlap", "summary": "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.", "body_md": "arXiv:2609.16071v1 Announce Type: new \nAbstract: 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.", "url": "https://wpnews.pro/news/schema-adaptive-action-conditioned-jepa-for-cross-machine-cnc-transfer-under", "canonical_source": "https://arxiv.org/abs/2609.16071", "published_at": "2026-09-16 04:00:00+00:00", "updated_at": "2026-09-16 04:07:05.043057+00:00", "lang": "en", "topics": ["machine-learning", "ai-research", "ai-infrastructure"], "entities": ["SAAC-JEPA", "Joint-Embedding Predictive Architecture", "PatchTST", "iTransformer", "RevIN", "arXiv"], "alternates": {"html": "https://wpnews.pro/news/schema-adaptive-action-conditioned-jepa-for-cross-machine-cnc-transfer-under", "markdown": "https://wpnews.pro/news/schema-adaptive-action-conditioned-jepa-for-cross-machine-cnc-transfer-under.md", "text": "https://wpnews.pro/news/schema-adaptive-action-conditioned-jepa-for-cross-machine-cnc-transfer-under.txt", "jsonld": "https://wpnews.pro/news/schema-adaptive-action-conditioned-jepa-for-cross-machine-cnc-transfer-under.jsonld"}}