Simulation-free Structure Learning for Stochastic Dynamics

2025-10-18

AI Summary

Existing methods struggle to jointly infer network structure and population-level dynamics from high-dimensional, stochastic, partially observed, and noisy physical systems—such as those in single-cell biology. This paper introduces StructureFlow, the first simulation-free joint modeling framework that unifies probabilistic flow modeling, causal inference, denoising diffusion, and variational inference to simultaneously learn both system structure and conditional dynamical trajectories directly from observational data. By overcoming the traditional decoupling of structural learning and dynamical modeling, StructureFlow achieves significant improvements in network recovery accuracy and trajectory inference fidelity across synthetic benchmarks, interpretable biological simulations, and real single-cell datasets. It establishes a novel paradigm for causal structure discovery and dynamic modeling under interventions.

Technology Category

• Machine Learning: Structured Learning
• Reasoning under Uncertainty: Relational Probabilistic Models
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Modeling dynamical systems and unraveling their underlying causal relationships is central to many domains in the natural sciences. Various physical systems, such as those arising in cell biology, are inherently high-dimensional and stochastic in nature, and admit only partial, noisy state measurements. This poses a significant challenge for addressing the problems of modeling the underlying dynamics and inferring the network structure of these systems. Existing methods are typically tailored either for structure learning or modeling dynamics at the population level, but are limited in their ability to address both problems together. In this work, we address both problems simultaneously: we present StructureFlow, a novel and principled simulation-free approach for jointly learning the structure and stochastic population dynamics of physical systems. We showcase the utility of StructureFlow for the tasks of structure learning from interventions and dynamical (trajectory) inference of conditional population dynamics. We empirically evaluate our approach on high-dimensional synthetic systems, a set of biologically plausible simulated systems, and an experimental single-cell dataset. We show that StructureFlow can learn the structure of underlying systems while simultaneously modeling their conditional population dynamics -- a key step toward the mechanistic understanding of systems behavior.

Problem

• Simultaneously learning structure and stochastic population dynamics
• Modeling high-dimensional systems with partial noisy measurements
• Addressing limitations in joint structure learning and dynamics modeling

Innovation

• Simulation-free joint structure and dynamics learning
• Handles high-dimensional stochastic physical systems
• Models conditional population dynamics from interventions