aiqbench analysis
This research introduces a provable pipeline using Observation Spectral Filtering (OSF) to distill nonlinear dynamical systems into compact linear state-space models, avoiding non-convex optimization pitfalls. The method guarantees that prediction error depends on observer complexity rather than latent dimension and demonstrates performance matching or exceeding directly trained baselines in MuJoCo behavior cloning tasks.
Spectral Distillation: From Nonlinear Dynamics to Linear State-Space Models
arXiv:2608.05416v1 Announce Type: new Abstract: Can nonlinear dynamical systems be learned through a compact linear state-space representation, without directly solving a non-convex system-identification problem? We give a provable pipeline for doing so. Starting from observations of an unknown nonlinear dynamical system, we first learn an implicit spectral predictor using Observation Spectral Filtering (OSF), a convex method that competes with the best linear observer for the system. We then apply spectral-to-LDS distillation to convert this predictor into an explicit recurrent linear dynamical system. Our main theorem shows that the average prediction error of the distilled LDS decomposes into an exponentially-small distillation term and the OSF learning term governed by the Luenberger complexity of the best observer. The guarantee is dimension-free: it depends on observer complexity rather than on the latent dimension needed to represent the nonlinear system. To our knowledge, this yields the first end-to-end provable method for extracting a best-in-hindsight LDS representation of nonlinear dynamics through convex learning followed by provable distillation. Experiments on linear LDS benchmarks and MuJoCo behavior cloning show that the train-then-distill pipeline produces compact LDS predictors that match or outperform directly trained baselines.
Abstract:Can nonlinear dynamical systems be learned through a compact linear state-space representation, without directly solving a non-convex system-identification problem? We give a provable pipeline for doing so. Starting from observations of an unknown nonlinear dynamical system, we first learn an implicit spectral predictor using Observation Spectral Filtering (OSF), a convex method that competes with the best linear observer for the system. We then apply spectral-to-LDS distillation to convert this predictor into an explicit recurrent linear dynamical system. Our main theorem shows that the average prediction error of the distilled LDS decomposes into an exponentially-small distillation term and the OSF learning term governed by the Luenberger complexity of the best observer. The guarantee is dimension-free: it depends on observer complexity rather than on the latent dimension needed to represent the nonlinear system. To our knowledge, this yields the first end-to-end provable method for extracting a best-in-hindsight LDS representation of nonlinear dynamics through convex learning followed by provable distillation. Experiments on linear LDS benchmarks and MuJoCo behavior cloning show that the train-then-distill pipeline produces compact LDS predictors that match or outperform directly trained baselines.
Machine Learning (cs.LG)
arXiv:2608.05416 [cs.LG]
(or arXiv:2608.05416v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2608.05416 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 276 words
Related coverage
Deep Spectral Learning of Embedded Latent Transfer Operators for Stochastic Dynamical Systems
arXiv · Jun 15, 2026
Ground-state preparation via nonlinear quantum dissipation
arXiv · Oct 8, 2026
Deep Embedded Multiplicative DMD for Algebra-Preserving Koopman Learning
arXiv · Jun 5, 2026
Koopman Theory and Metaethics
LessWrong · Sep 19, 2026