Beyond Domain-Specific World Models: JEPA-Anything Uses 1 Recipe for 7 Fields
JEPA-Anything factorizes latent targets into four predictors and beats JEPA baselines across seven domains.
MarkTechPost reports JEPA-Anything, a world-model recipe that splits a JEPA's single latent target into four orthogonal factors, each with its own predictor. Tested across seven domains, it beat matched JEPA baselines on all 10 dynamics tasks. On Interventional Pong, intervention error fell by 34.8%.
- Splits one JEPA latent target into four orthogonal factors
- Each factor uses its own predictor
- Beat matched JEPA baselines on all 10 dynamics tasks
- Cut Interventional Pong intervention error by 34.8%
- One recipe evaluated across seven domains
JEPA-Anything splits a JEPA's single latent target into 4 orthogonal factors, each with its own predictor. Tested across 7 domains, it beat matched JEPA baselines on all 10 dynamics tasks and cut Interventional Pong intervention error by 34.8%. The post Beyond Domain-Specific World Models: JEPA-Anything Uses 1 Recipe for 7 Fields appeared first on MarkTechPost.
The full text could not be extracted from this site (paywall, bot protection or heavy scripting). Read it at marktechpost.com.