AnisoWM: anisotropic representations improve planning in JEPA world models
SeoulNatlUniv · hf · 2026-09-30
Seoul National University researchers propose AnisoWM with ΛReg, fixing a geometry mismatch in latent world model planning.
- Finding: accurate prediction and noncollapsed representations don't guarantee a task-aligned latent planning cost — isotropic Gaussian regularization induces a geometry that ranks feasible outcomes differently from the true task cost.
- Method: replace the fixed isotropic Gaussian target with a learnable diagonal covariance under fixed-trace and anisotropy constraints; prediction objective, predictor architecture, and Euclidean planner stay unchanged, with the target used only in training.
- Results: across four visual control environments, AnisoWM beats LeWorldModel in planning success in all four, with better agreement between latent planning cost and task outcomes.
More from Research
- 16 Mathematicians Publish Leiden Declaration on AI's Role in Mathematics Research — burny_tech · 2026-09-30
- Repeated Solutions Make Reasoning Fragile After Instruction Tuning, Paper Finds — burny_tech · 2026-09-30
- AI is the discipline of building minds, and mathematics is just getting started — burny_tech · 2026-09-30
- A new auto-research loop that bootstraps the shape of the best possible result — burny_tech · 2026-09-30
- EasyPPO: just fix the critic — stable PPO for LLM post-training with zero training collapse — teortaxesTex · 2026-09-30
- Looped MoE Tuning: 2x Experts, 0.5x Looped Layers, 2x Loops, Attention Untied — burny_tech · 2026-09-30