BeliefGraph-JEPA: Structured Latent World Models for Action-Conditioned Time Series

2026-10-04

AI Summary

This study addresses the challenge of explicitly modeling how future driving factors influence multiple targets in action-conditioned temporal prediction. We propose a structured latent world model built upon the Joint Embedding Predictive Architecture (JEPA), introducing factorized latent effect states and a graph routing mechanism for the first time. Specifically, driving influences are decomposed into evolving latent states and precisely routed to target nodes via a graph network, supplemented by capacity-controlled residual connections that convey direct information. This design effectively decouples dynamic evolution from target-specific effects, enabling interpretable predictions. Experiments across four complex systems, including clinical settings, demonstrate that our model significantly outperforms existing baselines. These results validate the effectiveness of latent state rollout and graph-prioritized residual routing in enhancing predictive accuracy.

Abstract

Action-conditioned time-series forecasting requires accounting for how future actions and exogenous forcings influence multiple targets through partially observed effects with different delays and persistence. Direct conditioning leaves the evolution and target-specific influence of these effects implicit in the predictor, while static relational graphs specify connections without tracking evolving effects. This motivates representing future-driver influence through structured latent states that evolve over the forecast horizon and route information to individual targets. We introduce BeliefGraph-JEPA, a structured latent world model that factorizes driver influence into typed latent-effect states. These states are rolled forward under future drivers and routed through a graph to target-specific nodes, forming the predictive base of a joint-embedding predictive architecture. A capacity-controlled residual supplements this base with direct driver information. On four multi-target clinical, agricultural, environmental, and industrial systems, the framework outperforms a range of pretrained and supervised known-future-covariate baselines. Matched controls isolate latent dynamics, future rollout, graph routing, and residual capacity; future rollout and graph-first residual routing improve forecasting across all four systems.

Problem

• Action-conditioned time-series forecasting
• Latent world models
• Multi-target prediction
• Exogenous forcings
• Structured latent states

Innovation

• Structured Latent World Model
• Joint-Embedding Predictive Architecture
• Action-Conditioned Time Series
• Graph Routing
• Typed Latent-Effect States