Latent Dynamics · 2026-10-06
BeliefGraph-JEPA: Structured Latent World Models for Action-Conditioned Time Series
Yue Li, Kangqi Ni, Zhen Tan, Tianlong Chen
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
BeliefGraph-JEPA is a structured latent world model for action-conditioned multi-target time-series forecasting. It factorizes future-driver influence into typed latent-effect states that are rolled forward under future drivers, routed through a typed graph to target-specific nodes, and used as the JEPA-space predictive base. A capacity-controlled gated residual injects direct driver information. The authors report improved forecasting over known-future-covariate baselines on four multi-target systems.
Why it matters
The authors target action-conditioned forecasting where driver effects are delayed, mechanism-specific, and shared across multiple targets, with partially observed influence. Their approach makes these structured, typed latent effects explicit—then uses future rollout plus graph-first routing as the predictive base—while isolating components via matched controls to show what drives improvements.
Method
- Factor future-driver influence into typed latent-effect states, roll them forward causally under future drivers, and route horizon-specific readouts through a relation-aware typed GNN to target nodes.
- Use future rollout (FR) where the graph trajectory and encoded future drivers jointly predict in JEPA space; designate graph-mediated trajectory as the predictive base (graph-first residual routing, FR-GR).
- Add a capacity-controlled gated residual adapter that injects direct driver information, constrained by a residual-magnitude penalty and graph-base supervision.
Limitation
The text does not state explicit limitations.
Abstract (from arXiv)
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.
Related papers
- ActTime: Action-Conditioned World Modeling for Time Series
- MDN-VJEPA: Multimodal Probabilistic Joint-Embedding World Models
- Does Joint-Embedding Predictive Architecture Pretraining Help Time Series Forecasting?
- JEPA-TTT: Persistent Test-Time Training of Latent World Models for Planning under Dynamics Shifts
- On the Divergence of Accuracy and Mechanism Consistency in Time Series World Models
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-04 · Indexed by Pub-AI: 2026-10-06 UTC
BeliefGraph-JEPA: Structured Latent World Models for Action-Conditioned Time Series
Preprint posted: 2026-10-04 · Original record ↗
- arxiv
- 2610.05409
Explicit source versions (1)
- v1 · 2026-10-04
- Preprint available
Source date: 2026-10-04 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
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.