Phys-JEPA: Physics-Informed Latent World Models for Multivariate Time-Series Forecasting

arXiv 2026 54.9 method

TLDR

Phys-JEPA introduces a physics-informed latent world model that decomposes predictive states into physical and residual components for multivariate time-series forecasting.

Reasoning

The paper presents a novel approach by enforcing physical consistency on latent states rather than decoded outputs, showing improvements on real-world datasets. However, the results are described as 'initial', and the evaluation is limited to three datasets without extensive ablation or comparison to state-of-the-art physics-informed methods.

Read-first score

Read-first score 54.9, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 21.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%

91.8

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.91830888

Methodology quality 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=baseline,dataset,metric,result

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code,dataset

Topical relevance 29%

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FoundationFrontierBridgeMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 437.

Keyword Scores

world model

world dynamics prediction

generative world model

world simulator

model-based reinforcement learning world model

interactive world model

video world model

Deep Analysis

Innovations

Decomposition of latent predictive states into physical and residual components
Imposing physical consistency directly on latent states and latent transitions rather than only on decoded forecasts
Using known physical variables to organize the representation space while retaining residual capacity for unresolved dynamics

Methodology

Phys-JEPA is a joint-embedding predictive architecture for multivariate time-series forecasting. It learns a latent world model where predictive states are decomposed into physical and residual components, and physical consistency is enforced on latent states and transitions. The model is evaluated on Jena Climate, Traffic, and Electricity datasets using MSE as the metric.

Key Results

On Jena Climate at H=24, aggregate MSE reduces from 0.12482 to 0.12273 and temperature MSE from 0.01892 to 0.01831. On Traffic, full Phys-JEPA improves aggregate MSE over the supervised baseline across all horizons, e.g., H=192 MSE from 0.800784 to 0.773873. On Electricity, the best variant depends on horizon: static latent consistency is strongest at H=24 and H=48, while full Phys-JEPA gives the best aggregate and target-variable MSE at H=192.

Limitations

Results are preliminary and not yet comprehensive across diverse domains
Best variant depends on horizon and dataset, requiring careful selection
Requires known physical variables to define physical components in the latent space

Resources

Paper