Fast and principled equation discovery from chaos to climate
Bayesian-ARGOS discovers governing equations from noisy chaos data 100x faster than rivals — climate modeling and control systems are next.
The Thesis
Bayesian-ARGOS is a hybrid frequentist-Bayesian sparse regression framework that extracts interpretable differential equations from scarce, noisy data across seven chaotic benchmarks, beating SINDy on data efficiency for all tested systems and noise tolerance on six of seven. The two-order-of-magnitude compute reduction over bootstrap-based ARGOS matters because it makes real-time or embedded equation discovery feasible, not just academic. The integration with SINDy-SHRED for high-dimensional sea surface temperature reconstruction hints at serious climate and geophysical applications.
Catalyst
Library-based sparse regression (SINDy, ARGOS) matured enough to expose a clear Pareto failure: speed vs. statistical rigor vs. automation couldn't coexist until someone combined frequentist screening as a fast pre-filter with targeted Bayesian inference only on the surviving candidates. The SINDy-SHRED representation learning pipeline, itself recent, created a concrete high-dimensional testbed that made the climate application tractable.
What's New
SINDy (Brunton et al. 2016) is fast but lacks principled uncertainty quantification and struggles with data scarcity; bootstrap-based ARGOS adds rigor but at prohibitive computational cost. Bayesian-ARGOS inserts a frequentist screening gate so Bayesian inference only runs on a small candidate set, cutting cost by ~100x without sacrificing the probabilistic diagnostics.
The Counter
The core claim — 100x faster than bootstrap ARGOS while matching or beating SINDy on noise tolerance — is compelling on paper, but the test suite is seven low-dimensional chaotic ODEs (Lorenz, Rössler variants and kin). That is exactly the regime where sparse regression already works well. The hard problems in climate science involve stochastic PDEs on irregular grids with missing data, regime shifts, and non-stationarity; none of those appear in the benchmarks. The SINDy-SHRED climate demo is encouraging but produces 'latent equations,' meaning you're discovering dynamics in a learned low-dimensional embedding, not the actual physical equations — interpretability is therefore indirect and model-dependent. More practically, SINDy already has a large installed base of practitioners and tooling; a new method needs to show a library, tutorials, and integration with existing pipelines before it displaces anything commercially. The cultural velocity score on this paper is literally zero, which tells you the ML and climate communities haven't picked this up yet.
Longs
- MSFT
- GOOGL
- NVDA
Shorts
- PySINDy maintainers / Brunton Lab ecosystem (methodology displacement if Bayesian-ARGOS ships a production library)
- Traditional numerical weather prediction vendors relying on hand-coded PDE solvers if data-driven equation discovery matures
- Proprietary physics simulation ISVs (Ansys, Dassault Systèmes) if interpretable learned equations substitute for expensive domain expert tuning
Enablers (Picks & Shovels)
- NVDA (GPU throughput for Bayesian posterior sampling at scale)
- AWS / GCP / Azure (cloud HPC for climate-scale runs)
- NumFOCUS / PyMC open-source ecosystem (Bayesian inference toolchain)
- ECMWF and NOAA (open reanalysis datasets enabling climate validation)
Private Watchlist
- Pasteur Labs
- Causality Link
- Relation Therapeutics
The Paper
Our ability to predict, control, and ultimately understand complex systems rests on discovering the equations that govern their dynamics. Identifying these equations directly from noisy, limited observations has therefore become a central challenge in data-driven science, yet existing library-based sparse regression methods force a compromise between automation, statistical rigor, and computational efficiency. Here we develop Bayesian-ARGOS, a hybrid framework that reconciles these demands by combining rapid frequentist screening with focused Bayesian inference, enabling automated equation discovery with principled uncertainty quantification at a fraction of the computational cost of existing methods. Tested on seven chaotic systems under varying data scarcity and noise levels, Bayesian-ARGOS outperforms two state-of-the-art methods in most scenarios. It surpasses SINDy in data efficiency for all systems and noise tolerance for six out of the seven, with a two-order-of-magnitude reduction in computational cost compared to bootstrap-based ARGOS. The probabilistic formulation additionally enables a suite of standard statistical diagnostics, including influence analysis and multicollinearity detection that expose failure modes otherwise opaque. When integrated with representation learning (SINDy-SHRED) for high dimensional sea surface temperature reconstruction, Bayesian-ARGOS increases the yield of valid latent equations with significantly improved long horizon stability. Bayesian-ARGOS thus provides a principled, automated, and computationally efficient route from scarce and noisy observations to interpretable governing equations, offering a practical framework for equation discovery across scales, from benchmark chaotic systems to the latent dynamics underlying global climate patterns.