Neuroscience

NeuroWorld: A Latent Brain World Model for Stimulus-Conditioned Human Brain Dynamics

Zijian Dong, Jianxiong Zhou, Kwun Kei Ng, Jan Paolo Macapinlac Balagtas, Zhizhou Li, Zijiao Chen, Juan Helen Zhou

Featured August 18, 2026

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content. Always refer to the original paper.

Simply

NeuroWorld learns how your brain's hidden states change based on what you see and hear, then uses this knowledge to accurately guess your future brain activity, without peeking ahead at what's coming.

In depth

The paper introduces NeuroWorld, a novel two-stage framework that models human brain activity as stimulus-conditioned evolution in a learned latent brain-state space. Crucially, it learns these dynamics through next-latent prediction without directly reconstructing fMRI signals, ensuring the latent representation is optimized for stable, causal future forecasting rather than current observation fitting.

Key Takeaways

1
NeuroWorld is the first brain world model to frame naturalistic fMRI prediction as causal state evolution in a learned latent space, moving beyond stimulus-to-response regression.
2
Its two-stage design, Latent Dynamics Learning (LDL) and Latent Rollout Decoding (LRD), disentangles endogenous brain states from exogenous stimuli and optimizes latent representations for transition-stable recursive prediction, not fMRI reconstruction.
3
The model achieves state-of-the-art multi-step rollout performance under strictly causal conditions across multiple benchmarks, including the newly introduced SG-MIND dataset, demonstrating robustness to long-horizon autoregressive drift.

Conceptual Flow

HIGH LEVEL

Methodology: Learning Brain Dynamics

The model learns how brain states change based on what you see and hear, then uses that knowledge to guess what your brain will do next.

Brain Scan Data

Stimulus Data

Learn Brain State Changes

Predicted Future Brain Activity

Results: Superior Future Prediction

Compared to older methods, NeuroWorld is much better at predicting how your brain will react to new things over time.

Old Prediction Methods

NeuroWorld Method

Compare Accuracy

Less Accurate Future Brain Activity

More Accurate Future Brain Activity

This breakdown was generated by SciGrove. Get the same analysis — intuition, storyboard, peer review, a runnable prototype and a glossary — on any paper you upload or paste a DOI for.