Early-Signal Detector

method · first seen 8 Oct 2026

surrogate model

Also: emulator, ML surrogate, surrogate modeling, surrogate modelling, surrogate model emulation, surrogate-based optimization, neural network surrogate model

A fast approximate model trained to emulate an expensive numerical simulator.

A surrogate model is a cheap learned approximation of an expensive simulator, used for optimization, uncertainty quantification, inverse design and scenario screening. The sources are a broad spread of domain applications rather than a new idea: transformer surrogates for nuclear severe-accident codes, wildfire hazard mapping across Canada, plasma fields compared across FNO and Transolver, landslide runout emulators, and a neural surrogate of NGSpice for cryogenic CMOS model extraction. The newer thread is testing pretrained time-series and tabular foundation models such as TabPFN as surrogates without task-specific training. The one GitHub hit is a game-console emulator, which is a mis-match.

Surrogate modelling has been standard in engineering and scientific computing for decades, so the rise here mostly reflects ordinary arXiv volume. The foundation-model-as-surrogate angle is the part worth watching separately.

Analyst note (established), 18 Nov 2026.

Mentions per week

0719 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Rank 41 on 15 Nov 2026 · 13 mentions in 28 days · 13 independent groups · papers → code · first seen 5 weeks ago · acceleration -0.02

Earliest sources

Latest sources

Mentioned together with

tabular foundation modeltransformer decoder surrogate for time seriesphysics-informed machine learningresidual learningBurnP3+Bayesian optimizationconditional generative adversarial networksoft modular robotsstellaratorU-Net