Latent Dynamics · 2026-10-02
Network World Models as Environments for Algorithm Design on Complex Systems
Rishab Alagharu, Hongji Pu, Zeeshan Memon, Xinyuan Song, Yuntong Hu, Liang Zhao
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
The authors propose an action-conditioned Network World Model that learns network diffusion dynamics under interventions over time, applies each intervention exactly, and predicts the outcome that follows. Combined with an LLM-based coding agent search loop that uses fast model rollouts plus rollout diagnostics/credit and counterfactual probes, they report matching or exceeding the strongest baselines in 138/141 settings and up to 14.5× faster rollouts than Monte Carlo simulation. Code will be released upon acceptance.
Why it matters
Evaluating candidate intervention algorithms on complex network diffusion tasks is expensive when done with Monte Carlo simulation. The paper provides a learned simulator that returns both fast outcome estimates and feedback for iterative algorithm design (full rollouts, action-level credit, and counterfactual probes), plus a regret bound linking performance to transition error, rollout horizon, and search suboptimality.
Method
- Action-conditioned Network World Model: exact intervention handling via a known T_exo, then learn one-step diffusion with a graph neural simulator and a process-constrained transition head.
- World-model rollout evaluator inside an LLM coding-agent loop: score candidate executable algorithms by rolling out the model forward over the task horizon under shared randomness.
- Use rollout diagnostics, action-level credit (leave-one-action-out under same rollout seed), and counterfactual probes (queries about alternative interventions) to revise code iteratively.
Limitation
The paper assumes the one-step transition error of the Network World Model is at most ε over the state-action support reachable by algorithms considered within horizon H, and uses this to bound algorithm-selection regret (Theorem 1).
Abstract (from arXiv)
World models, which simulate an environment and predict how it changes under actions, are increasingly used in real-world applications such as robotics. Complex systems call for the same tool because the effect of an action is not immediate. Seeding nodes for a campaign, or immunizing nodes against an epidemic, changes little on its own; what matters is the outcome that unfolds over the steps that follow. Designing an algorithm that selects such actions to maximize expected performance on a task is inherently iterative, and every candidate must be scored by the outcome it produces. Obtaining that outcome has relied on simulation, whose cost becomes a bottleneck when candidates are evaluated over many sampled trajectories. We propose an action-conditioned Network World Model that learns a network's diffusion dynamics under interventions over time, applies each action to the network, and predicts the outcome that follows. It serves as a fast evaluator inside an algorithm design loop in which a coding agent designs and refines executable algorithms using feedback from full rollouts, action-level credit, and counterfactual probes over alternative interventions. Across eight network tasks and five diffusion models, the designed algorithms match or exceed the strongest reported baseline in 138 of 141 settings while enabling up to 14.5 times faster rollouts than Monte Carlo simulation. Code will be released upon acceptance.
Related papers
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-01 · Indexed by Pub-AI: 2026-10-02 UTC
Network World Models as Environments for Algorithm Design on Complex Systems
Preprint posted: 2026-10-01 · Original record ↗
- arxiv
- 2610.01048
Explicit source versions (1)
- v1 · 2026-10-01
- Preprint available
Source date: 2026-10-01 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
World models, which simulate an environment and predict how it changes under actions, are increasingly used in real-world applications such as robotics. Complex systems call for the same tool because the effect of an action is not immediate. Seeding nodes for a campaign, or immunizing nodes against an epidemic, changes little on its own; what matters is the outcome that unfolds over the steps that follow. Designing an algorithm that selects such actions to maximize expected performance on a task is inherently iterative, and every candidate must be scored by the outcome it produces. Obtaining that outcome has relied on simulation, whose cost becomes a bottleneck when candidates are evaluated over many sampled trajectories. We propose an action-conditioned Network World Model that learns a network's diffusion dynamics under interventions over time, applies each action to the network, and predicts the outcome that follows. It serves as a fast evaluator inside an algorithm design loop in which a coding agent designs and refines executable algorithms using feedback from full rollouts, action-level credit, and counterfactual probes over alternative interventions. Across eight network tasks and five diffusion models, the designed algorithms match or exceed the strongest reported baseline in 138 of 141 settings while enabling up to 14.5 times faster rollouts than Monte Carlo simulation. Code will be released upon acceptance.