Rollout-Decoded Reconstruction for Long-Horizon Prediction in Latent World Models
cs.LG
Submitted: 2026-08-25
Updated: 2026-08-25
Comments: 15 pages, 6 figures
License: http://creativecommons.org/licenses/by/4.0/
The gist: A latent world model trains its decoder on latents anchored to observations, then deploys it on the model's own free-running rollout, hundreds of steps past the last observation.
Terminology
Abstract
A latent world model trains its decoder on latents anchored to observations, then deploys it on the model's own free-running rollout, hundreds of steps past the last observation. Rollout-Decoded Reconstruction (RDR) closes this gap with a single loss term that free-runs the model during training exactly as evaluation will, decodes every rollout latent, and penalizes reconstruction error against ground truth. The term adds no parameters, costs training-time compute only, and reduces to the standard objective at weight zero, so every comparison in this paper is a one-flag A/B. On the chaotic Kuramoto-Sivashinsky equation, RDR raises valid prediction time (the time to first crossing of normalized error 0.5) from 3.87 plus or minus 0.23 to 6.97 plus or minus 0.42 time units at an identical 193,568 parameters, a 1.80 times improvement confirmed on seeds never used in selection and holding in 10 of 10 preregistered configurations at ratios of 1.71-2.50 times. The results come from a single system; a sweep in which the advantage grows with latent width is descriptive, and control experiments on two classic tasks are preliminary.
Sources
- Scheduled Sampling for Sequence Prediction with Recurrent Neural Networks
- Message Passing Neural PDE Solvers
- MuDreamer: Learning Predictive World Models without Reconstruction
- Diffusion Forcing: Next-token Prediction Meets Full-Sequence Diffusion
- Deep Reinforcement Learning in a Handful of Trials using Probabilistic Dynamics Models
- NeuroWorld: A Latent Brain World Model for Stimulus-Conditioned Human Brain Dynamics
- Lyapunov exponents of the Kuramoto-Sivashinsky PDE
- Dream to Control: Learning Behaviors by Latent Imagination
- Learning Latent Dynamics for Planning from Pixels
- Mastering Atari with Discrete World Models
- Mastering Diverse Domains through World Models
- TD-MPC2: Scalable, Robust World Models for Continuous Control
- Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion
- Professor Forcing: A New Algorithm for Training Recurrent Networks
- Koopman Dreamer: Spectrally Constrained Latent Dynamics for Stable World-Model Imagination
- Deep learning to discover and predict dynamics on an inertial manifold
- Sample-efficient Cross-Entropy Method for Real-time Planning
- Simplified State Space Layers for Sequence Modeling
- Rollout-LaSDI: Enhancing the long-term accuracy of Latent Space Dynamics
- Solver-in-the-Loop: Learning from Differentiable Physics to Interact with Iterative PDE-Solvers
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: Local-Consistency Optimisation for Bayesian Neural Networks
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks