![[2510.26656] Heuristic Adaptation of Potentially Misspecified Domain Support for Likelihood-Free Inference in Stochastic Dynamical Systems](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2510.26656] Heuristic Adaptation of Potentially Misspecified Domain Support for Likelihood-Free Inference in Stochastic Dynamical Systems
Summary
This article presents a novel approach to likelihood-free inference (LFI) in robotics, addressing the issue of potentially misspecified domain support and proposing three heuristic adaptations to improve inference accuracy.
Why It Matters
The findings are significant for the field of robotics as they enhance the reliability of inference methods in stochastic dynamical systems. By addressing support misspecification, the proposed heuristics can lead to better parameter inference and improved performance in robotic tasks, which is crucial for developing more autonomous and capable robotic systems.
Key Takeaways
- Introduces three heuristic LFI variants: EDGE, MODE, and CENTRE.
- Addresses the issue of support misspecification in likelihood-free inference.
- Demonstrates improved parameter inference and policy learning in dynamic tasks.
Computer Science > Robotics arXiv:2510.26656 (cs) [Submitted on 30 Oct 2025 (v1), last revised 25 Feb 2026 (this version, v3)] Title:Heuristic Adaptation of Potentially Misspecified Domain Support for Likelihood-Free Inference in Stochastic Dynamical Systems Authors:Georgios Kamaras, Craig Innes, Subramanian Ramamoorthy View a PDF of the paper titled Heuristic Adaptation of Potentially Misspecified Domain Support for Likelihood-Free Inference in Stochastic Dynamical Systems, by Georgios Kamaras and 2 other authors View PDF HTML (experimental) Abstract:In robotics, likelihood-free inference (LFI) can provide the domain distribution that adapts a learnt agent in a parametric set of deployment conditions. LFI assumes an arbitrary support for sampling, which remains constant as the initial generic prior is iteratively refined to more descriptive posteriors. However, a potentially misspecified support can lead to suboptimal, yet falsely certain, posteriors. To address this issue, we propose three heuristic LFI variants: EDGE, MODE, and CENTRE. Each interprets the posterior mode shift over inference steps in its own way and, when integrated into an LFI step, adapts the support alongside posterior inference. We first expose the support misspecification issue and evaluate our heuristics using stochastic dynamical benchmarks. We then evaluate the impact of heuristic support adaptation on parameter inference and policy learning for a dynamic deformable linear object (DLO) manipulati...
