![[2602.13813] Pawsterior: Variational Flow Matching for Structured Simulation-Based Inference](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.13813] Pawsterior: Variational Flow Matching for Structured Simulation-Based Inference
Summary
Pawsterior introduces a variational flow-matching framework to enhance simulation-based inference (SBI), addressing constraints in structured domains and discrete latent structures.
Why It Matters
This research is significant as it expands the capabilities of simulation-based inference methods, allowing for more accurate modeling in complex domains. By incorporating geometric constraints and handling discrete variables, it opens new avenues for applications in machine learning and artificial intelligence.
Key Takeaways
- Pawsterior improves simulation-based inference by addressing geometric constraints.
- The framework allows for better handling of discrete latent structures.
- It enhances numerical stability and posterior fidelity in SBI tasks.
- The method generalizes existing flow-matching techniques for structured domains.
- Pawsterior enables access to previously inaccessible SBI problems.
Computer Science > Machine Learning arXiv:2602.13813 (cs) [Submitted on 14 Feb 2026] Title:Pawsterior: Variational Flow Matching for Structured Simulation-Based Inference Authors:Jorge Carrasco-Pollo, Floor Eijkelboom, Jan-Willem van de Meent View a PDF of the paper titled Pawsterior: Variational Flow Matching for Structured Simulation-Based Inference, by Jorge Carrasco-Pollo and 1 other authors View PDF Abstract:We introduce Pawsterior, a variational flow-matching framework for improved and extended simulation-based inference (SBI). Many SBI problems involve posteriors constrained by structured domains, such as bounded physical parameters or hybrid discrete-continuous variables, yet standard flow-matching methods typically operate in unconstrained spaces. This mismatch leads to inefficient learning and difficulty respecting physical constraints. Our contributions are twofold. First, generalizing the geometric inductive bias of CatFlow, we formalize endpoint-induced affine geometric confinement, a principle that incorporates domain geometry directly into the inference process via a two-sided variational model. This formulation improves numerical stability during sampling and leads to consistently better posterior fidelity, as demonstrated by improved classifier two-sample test performance across standard SBI benchmarks. Second, and more importantly, our variational parameterization enables SBI tasks involving discrete latent structure (e.g., switching systems) that are fun...