arxiv: 2605.00510 · v1 · submitted 2026-05-01 · 💻 cs.LG · cs.CV· physics.comp-ph

Recognition: unknown

Scale-Aware Adversarial Analysis: A Diagnostic for Generative AI in Multiscale Complex Systems

Duo Xu, Guang-Xing Li, Keping Qiu, Mengke Zhao

Pith reviewed 2026-05-09 19:52 UTC · model grok-4.3

classification 💻 cs.LG cs.CVphysics.comp-ph

keywords generative modelsmultiscale systemsdiffusion decompositionscale-aware analysisphysical constraintsmodel instabilityadversarial diagnostics

0 comments

The pith

Generative models for multiscale physics exhibit localized structural freezing and nonlinear divergence under scale-aware perturbations

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces Constrained Diffusion Decomposition as a diagnostic to test whether generative models internalize the continuous multiscale dynamics of physical systems or only interpolate discrete statistics. When applied to a denoising diffusion probabilistic model, it shows that the network produces localized freezing and instability instead of smooth PDE-like responses when subjected to physically constrained scale modifications. A reader would care because reliable simulation of systems like turbulence or cosmic structure formation requires models that respect cross-scale causality rather than drifting into unphysical states. The framework supplies a controlled method for generating coherent test states to expose these algorithmic limits.

Core claim

Under moderate physical perturbations executed via deterministic interventions in the continuous CDD-based scale space, the unconstrained generative model exhibits localized structural freezing and non-linear instability rather than continuous PDE-like responses. The network fails to maintain cross-scale continuity, causing the generative trajectory to diverge when pushed into unseen physical states.

What carries the argument

Constrained Diffusion Decomposition (CDD), a diffusion-based multiscale data decomposition algorithm that performs physically constrained data generation and model evaluation through scale-aware modifications

If this is right

Generative models must incorporate mechanisms that enforce cross-scale continuity when modeling multiscale physical systems.
The CDD framework supplies a testbed for exposing algorithmic vulnerabilities before deployment in scientific simulations.
Future architectures will need explicit physical constraints to respect multiscale causality in the natural universe.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

This diagnostic approach could be adapted to other data-driven models to check whether they preserve continuous dynamics beyond the training distribution.
Standard pixel-wise perturbation methods in explainable AI are likely to remain inadequate for any system whose governing laws operate across continuous scales.

Load-bearing premise

Modifications produced by Constrained Diffusion Decomposition remain inside the valid physical distribution and represent meaningful real-world multiscale perturbations without introducing unphysical artifacts.

What would settle it

A generative model that maintains continuous cross-scale responses and PDE-like behavior under identical CDD perturbations would falsify the claim of inherent structural freezing and divergence.

Figures

Figures reproduced from arXiv: 2605.00510 by Duo Xu, Guang-Xing Li, Keping Qiu, Mengke Zhao.

**Figure 2.** Figure 2: The Negative Response Paradox and Violation of structural Monotonicity. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Figure 3: Amplitude Diagnostics: Topological Freezing and Prior-Anchored Hallucinations. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

**Figure 4.** Figure 4: Geometric Diagnosis of Score Field Divergence and Manifold Fragmentation. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

**Figure 5.** Figure 5: Mathematical formulation and morphological manifestation of multiscale coherent modification. (a) Geometric Rotation in CDD-based Scale Space: The scale-component density ni versus spatial scale ri is plotted in the density-scale phase space. The augmentation pipeline (Equation 3) rotates the natural scaling cascade (κρ = −2.78) around a macroscopic boundary condition (rref ). Within the physically permiss… view at source ↗

read the original abstract

Complex physical systems, from supersonic turbulence to the macroscopic structure of the universe, are governed by continuous multiscale dynamics. While modern machine learning architectures excel at mapping the high-dimensional observables of these systems, it remains unclear whether they internalize the governing physical laws or merely interpolate discrete statistical correlations. Standard Explainable AI (XAI) architectures, particularly perturbation-based and gradient-saliency methods, rely on pixel-wise perturbations, which generate unphysical artifacts and push inputs off the valid empirical distribution. To resolve this, we introduce a diagnostic framework driven by Constrained Diffusion Decomposition (CDD), a diffusion-based multiscale data decomposition algorithm that enables physically constrained data generation and model evaluation via scale-aware modifications. Applying this framework to a Denoising Diffusion Probabilistic Model (DDPM), we execute deterministic interventions directly within the continuous, CDD-based scale space. We demonstrate that under moderate physical perturbations, the unconstrained generative model exhibits localized structural freezing and non-linear instability rather than continuous PDE-like responses. The network fails to maintain cross-scale continuity, causing the generative trajectory to diverge when pushed into unseen physical states. By synthesizing a continuum of physically coherent states, this scale-informed methodology establishes a controlled test ground to evaluate algorithmic vulnerabilities, providing the rigorous physical constraints necessary for future architectures to respect the multiscale causality of the natural universe.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper's new diagnostic using constrained diffusion decomposition to probe cross-scale continuity in DDPMs is a reasonable idea, but the abstract gives no evidence that the perturbations are physically valid.

read the letter

The main thing here is a new way to test generative models on multiscale physical data by decomposing inputs with a diffusion process that keeps changes scale-aware and supposedly constrained. They apply it to a DDPM and report that the model shows structural freezing and diverges under moderate perturbations instead of behaving like a continuous PDE solver. That points to a real gap in how these networks handle cross-scale causality, which matters for applications like turbulence or cosmology simulations. The approach improves on standard XAI perturbations that create obvious unphysical artifacts, and framing the test inside a continuous scale space is a clean move. Credit for trying to tie the evaluation directly to the multiscale nature of the target systems rather than generic pixel noise. The soft spot is that nothing in the available text shows the CDD outputs actually respect conservation laws, dispersion relations, or match numerical PDE solutions. The central claim about non-linear instability and loss of continuity rests on those perturbations being meaningful physical interventions; if they drift off the valid manifold, the divergence is just an artifact of the test. No quantitative metrics, error bars, or method details appear, so the demonstration stays at the level of a high-level observation. This is aimed at researchers building generative models for scientific data who already care about physical consistency. A reader working on scientific machine learning could get value from the framing even if the results need more backing. It deserves a serious referee because the problem it targets is genuine and the proposed fix is specific enough to evaluate, though the current version looks preliminary and would likely come back with requests for validation experiments against known physics.

Referee Report

1 major / 1 minor

Summary. The paper introduces Constrained Diffusion Decomposition (CDD), a diffusion-based multiscale data decomposition algorithm, to enable physically constrained, scale-aware perturbations for diagnosing whether generative models such as Denoising Diffusion Probabilistic Models (DDPM) internalize continuous multiscale physical dynamics rather than discrete statistical correlations. Applying deterministic interventions in CDD scale space to an unconstrained DDPM, the work claims to show localized structural freezing and non-linear instability instead of continuous PDE-like responses, with the generative trajectory diverging under unseen physical states due to failure to maintain cross-scale continuity.

Significance. If the central demonstration holds, the framework supplies a controlled, continuum-based testbed for exposing vulnerabilities in generative AI applied to multiscale systems (e.g., turbulence or cosmology), crediting the synthesis of physically coherent states via CDD as a step toward rigorous physical constraints on model evaluation. This could inform development of architectures that better respect multiscale causality.

major comments (1)

Abstract: the claim that CDD enables 'physically constrained data generation' and 'scale-aware modifications' that remain inside the valid physical distribution is asserted without any reported check against governing equations, conservation laws, dispersion relations, or consistency with numerical PDE solutions. Because every reported trajectory, instability metric, and divergence observation depends on this assumption, the demonstration that the DDPM exhibits structural freezing rather than continuous responses cannot be verified as evidence of model failure versus an artifact of off-manifold test states.

minor comments (1)

Abstract: quantitative results, error bars, specific instability metrics, and details of the DDPM architecture or intervention magnitudes are absent, limiting immediate assessment of effect sizes.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading and for identifying a key gap in the validation of our claims. We address the major comment below.

read point-by-point responses

Referee: Abstract: the claim that CDD enables 'physically constrained data generation' and 'scale-aware modifications' that remain inside the valid physical distribution is asserted without any reported check against governing equations, conservation laws, dispersion relations, or consistency with numerical PDE solutions. Because every reported trajectory, instability metric, and divergence observation depends on this assumption, the demonstration that the DDPM exhibits structural freezing rather than continuous responses cannot be verified as evidence of model failure versus an artifact of off-manifold test states.

Authors: We agree that the abstract asserts physical fidelity of CDD without explicit verification against governing equations or conservation laws. The current manuscript relies on the construction of CDD (diffusion decomposition constrained to the learned data manifold) but does not report direct comparisons to PDE solutions or dispersion relations. We will revise the manuscript to include such checks on the turbulence and cosmology datasets, for example by quantifying conservation errors and consistency with numerical solvers for the perturbed states. This will allow readers to confirm that the reported instabilities reflect model behavior rather than off-manifold artifacts. We will also moderate the abstract language pending these additions. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical diagnostic stands independent of inputs

full rationale

The paper introduces CDD as a novel decomposition method and applies it to observe DDPM behavior under scale-aware perturbations. The reported findings (localized freezing, non-linear instability, divergence from PDE-like continuity) are presented as direct empirical outcomes of those interventions rather than quantities derived by fitting or self-definition. No equations reduce a prediction to a fitted parameter, no uniqueness theorem is imported from self-citation, and the framework's physical-constraint claim is definitional to the new tool rather than a tautological restatement of the model evaluation results. The derivation chain is therefore self-contained observational analysis.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The framework rests on the assumption that diffusion-based decomposition can isolate physically meaningful scales and that perturbations applied in this space test genuine model understanding rather than artifacts.

axioms (1)

domain assumption Multiscale physical systems admit a continuous decomposition into scale components that can be modified independently while remaining on the valid data manifold.
Invoked to justify the use of CDD for physically constrained interventions.

invented entities (1)

Constrained Diffusion Decomposition (CDD) no independent evidence
purpose: To perform multiscale data decomposition enabling physically constrained generation and evaluation.
New algorithm introduced to address limitations of pixel-wise perturbation methods.

pith-pipeline@v0.9.0 · 5545 in / 1236 out tokens · 46242 ms · 2026-05-09T19:52:38.351653+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

115 extracted references · 101 canonical work pages · 4 internal anchors

[1]

apj , keywords =

Changes of Dust Opacity with Density in the Orion A Molecular Cloud. apj , keywords =. doi:10.1088/0004-637X/763/1/55 , archivePrefix =. 1211.6475 , primaryClass =

work page doi:10.1088/0004-637x/763/1/55
[2]

apjl , keywords =

Two Mass Distributions in the L 1641 Molecular Clouds: The Herschel Connection of Dense Cores and Filaments in Orion A. apjl , keywords =. doi:10.1088/2041-8205/777/2/L33 , archivePrefix =. 1309.2332 , primaryClass =

work page doi:10.1088/2041-8205/777/2/l33 2041
[3]

, keywords =

The Photodetector Array Camera and Spectrometer (PACS) on the Herschel Space Observatory. , keywords =. doi:10.1051/0004-6361/201014535 , archivePrefix =. 1005.1487 , primaryClass =

work page doi:10.1051/0004-6361/201014535
[4]

J., Abergel, A., Abreu, A., et al

The Herschel-SPIRE instrument and its in-flight performance. , keywords =. doi:10.1051/0004-6361/201014519 , archivePrefix =. 1005.5123 , primaryClass =

work page doi:10.1051/0004-6361/201014519
[5]

mnras , keywords =

Large-scale filaments associated with Milky Way spiral arms. mnras , keywords =. doi:10.1093/mnras/stv735 , archivePrefix =. 1504.00647 , primaryClass =

work page doi:10.1093/mnras/stv735
[6]

, keywords =

Turbulence and star formation in molecular clouds. mnras , keywords =. doi:10.1093/mnras/194.4.809 , adsurl =

work page doi:10.1093/mnras/194.4.809
[7]

, keywords =

Gravitational collapse of the OMC-1 region. , keywords =. doi:10.1051/0004-6361/201730732 , archivePrefix =. 1703.03464 , primaryClass =

work page doi:10.1051/0004-6361/201730732
[8]

mnras , keywords =

Density exponent analysis: gravity-driven steepening of the density profiles of star-forming regions. mnras , keywords =. doi:10.1093/mnrasl/slac049 , archivePrefix =. 2205.03215 , primaryClass =

work page doi:10.1093/mnrasl/slac049
[9]

F., & Ostriker, E

Theory of Star Formation. araa , keywords =. doi:10.1146/annurev.astro.45.051806.110602 , archivePrefix =. 0707.3514 , primaryClass =

work page doi:10.1146/annurev.astro.45.051806.110602
[10]

2012, , 50, 29, 10.1146/annurev-astro-081811-125514

Magnetic Fields in Molecular Clouds. araa , year = 2012, month = sep, volume =. doi:10.1146/annurev-astro-081811-125514 , adsurl =

work page doi:10.1146/annurev-astro-081811-125514 2012
[11]

Turbulence in Milky Way Star-forming Regions Traced by Young Stars and Gas.ApJ2022,934, 7, [arXiv:astro-ph.GA/2205.00012]

Turbulence in Milky Way Star-forming Regions Traced by Young Stars and Gas. apj , keywords =. doi:10.3847/1538-4357/ac76bf , archivePrefix =. 2205.00012 , primaryClass =

work page doi:10.3847/1538-4357/ac76bf
[12]

mnras , keywords =

Criteria for gravitational instability and quasi-isolated gravitational collapse in turbulent medium. mnras , keywords =. doi:10.1093/mnras/stw2707 , archivePrefix =. 1610.06577 , primaryClass =

work page doi:10.1093/mnras/stw2707
[13]

mnras , keywords =

Scale-free gravitational collapse as the origin of r ^ -2 density profile - a possible role of turbulence in regulating gravitational collapse. mnras , keywords =. doi:10.1093/mnras/sty657 , archivePrefix =. 1803.03273 , primaryClass =

work page doi:10.1093/mnras/sty657
[14]

, keywords =

Hierarchical star cluster assembly in globally collapsing molecular clouds. mnras , keywords =. doi:10.1093/mnras/stw3229 , archivePrefix =. 1611.00088 , primaryClass =

work page doi:10.1093/mnras/stw3229
[15]

Towards a comprehensive scenario

Global hierarchical collapse in molecular clouds. Towards a comprehensive scenario. mnras , keywords =. doi:10.1093/mnras/stz2736 , archivePrefix =. 1903.11247 , primaryClass =

work page doi:10.1093/mnras/stz2736 1903
[16]

mnras , keywords =

Multiscale accretion in dense cloud cores and the delayed formation of massive stars. mnras , keywords =. doi:10.1093/mnras/stae1090 , archivePrefix =. 2306.13846 , primaryClass =

work page doi:10.1093/mnras/stae1090
[17]

, keywords =

Gravity or turbulence? - II. Evolving column density probability distribution functions in molecular clouds. mnras , keywords =. doi:10.1111/j.1365-2966.2011.19141.x , archivePrefix =. 1105.5411 , primaryClass =

work page doi:10.1111/j.1365-2966.2011.19141.x 2011
[18]

apj , keywords =

The Star Formation Rate in the Gravoturbulent Interstellar Medium. apj , keywords =. doi:10.3847/1538-4357/aad002 , archivePrefix =. 1801.05428 , primaryClass =

work page doi:10.3847/1538-4357/aad002
[19]

mnras , keywords =

On the shape and completeness of the column density probability distribution function of molecular clouds. mnras , keywords =. doi:10.1093/mnras/sty3071 , archivePrefix =. 1811.02864 , primaryClass =

work page doi:10.1093/mnras/sty3071
[20]

, keywords =

Evolution of column density distributions within Orion A. , keywords =. doi:10.1051/0004-6361/201526243 , archivePrefix =. 1504.05188 , primaryClass =

work page doi:10.1051/0004-6361/201526243
[21]

2010, ApJL, 723, L7, doi: 10.1088/2041-8205/723/1/L7

How Many Infrared Dark Clouds Can form Massive Stars and Clusters?. apjl , keywords =. doi:10.1088/2041-8205/723/1/L7 , archivePrefix =. 1009.1617 , primaryClass =

work page doi:10.1088/2041-8205/723/1/l7 2041
[22]

, keywords =

Density profile evolution during prestellar core collapse: collapse starts at the large scale. mnras , keywords =. doi:10.1093/mnras/stab394 , archivePrefix =. 2009.14151 , primaryClass =

work page doi:10.1093/mnras/stab394 2009
[23]

, keywords =

Star formation in molecular clouds: observation and theory. araa , keywords =. doi:10.1146/annurev.aa.25.090187.000323 , adsurl =

work page doi:10.1146/annurev.aa.25.090187.000323
[24]

E., Heitsch, F., et al

Initial Conditions for Star Formation: a Physical Description of the Filamentary ISM. Protostars and Planets VII , year = 2023, editor =. doi:10.48550/arXiv.2203.09562 , archivePrefix =. 2203.09562 , primaryClass =

work page doi:10.48550/arxiv.2203.09562 2023
[25]

Galaxies , keywords =

Magnetic Fields in Molecular Clouds Observation and Interpretation. Galaxies , keywords =. doi:10.3390/galaxies9020041 , archivePrefix =. 2106.08172 , primaryClass =

work page doi:10.3390/galaxies9020041
[26]

apj , keywords =

On the Evolution of the Density Probability Density Function in Strongly Self-gravitating Systems. apj , keywords =. doi:10.1088/0004-637X/781/2/91 , archivePrefix =. 1310.4346 , primaryClass =

work page doi:10.1088/0004-637x/781/2/91
[27]

Philosophical Transactions of the Royal Society of London Series A , year = 1902, month = jan, volume =

The Stability of a Spherical Nebula. Philosophical Transactions of the Royal Society of London Series A , year = 1902, month = jan, volume =. doi:10.1098/rsta.1902.0012 , adsurl =

work page doi:10.1098/rsta.1902.0012 1902
[28]

arXiv e-prints , keywords =

Super-Jeans fragmentation in massive star-forming regions revealed by triangulation analysis. arXiv e-prints , keywords =. doi:10.48550/arXiv.2410.14610 , archivePrefix =. 2410.14610 , primaryClass =

work page doi:10.48550/arxiv.2410.14610
[29]

and Haehnelt, M

Helical fields and filamentary molecular clouds - I. mnras , keywords =. doi:10.1046/j.1365-8711.2000.03066.x , archivePrefix =. astro-ph/9901096 , primaryClass =

work page doi:10.1046/j.1365-8711.2000.03066.x 2000
[30]

2023, MNRAS, 520, 3259, doi: 10.1093/mnras/stad012

ATOMS: ALMA Three-millimeter Observations of Massive Star-forming regions - XV. Steady accretion from global collapse to core feeding in massive hub-filament system SDC335. mnras , keywords =. doi:10.1093/mnras/stad012 , archivePrefix =. 2301.01895 , primaryClass =

work page doi:10.1093/mnras/stad012
[31]

, keywords =

The distance to the Orion Nebula. , keywords =. doi:10.1051/0004-6361:20078247 , archivePrefix =. 0709.0485 , primaryClass =

work page doi:10.1051/0004-6361:20078247
[32]

apj , keywords =

The Two States of Star-forming Clouds. apj , keywords =. doi:10.1088/0004-637X/750/1/13 , archivePrefix =. 1202.2594 , primaryClass =

work page doi:10.1088/0004-637x/750/1/13
[33]

C., & Lazarian, A

Observational Diagnostics of Self-gravitating MHD Turbulence in Giant Molecular Clouds. apj , keywords =. doi:10.1088/0004-637X/808/1/48 , archivePrefix =. 1505.03855 , primaryClass =

work page doi:10.1088/0004-637x/808/1/48
[34]

apj , keywords =

Slope of Magnetic Field Density Relation as an Indicator of Magnetic Dominance. apj , keywords =. doi:10.3847/1538-4357/ad8b4d , archivePrefix =. 2409.02786 , primaryClass =

work page doi:10.3847/1538-4357/ad8b4d
[35]

apjs , keywords =

Multiscale Decomposition of Astronomical Maps: A Constrained Diffusion Method. apjs , keywords =. doi:10.3847/1538-4365/ac4bc4 , archivePrefix =. 2201.05484 , primaryClass =

work page doi:10.3847/1538-4365/ac4bc4
[36]

apj , keywords =

The Catalogue for Astrophysical Turbulence Simulations (CATS). apj , keywords =. doi:10.3847/1538-4357/abc484 , archivePrefix =. 2010.11227 , primaryClass =

work page doi:10.3847/1538-4357/abc484 2010
[37]

apjs , keywords =

Cosmological Adaptive Mesh Refinement Magnetohydrodynamics with Enzo. apjs , keywords =. doi:10.1088/0067-0049/186/2/308 , archivePrefix =. 0902.2594 , primaryClass =

work page doi:10.1088/0067-0049/186/2/308
[38]

Bird's eye view of molecular clouds in the Milky Way. I. Column density and star formation from sub-parsec to kiloparsec scales. , keywords =. doi:10.1051/0004-6361/202040021 , archivePrefix =. 2108.04518 , primaryClass =

work page doi:10.1051/0004-6361/202040021
[39]

Magnetic Field of Molecular Gas Measured with the Velocity Gradient Technique I. Orion A. apj , keywords =. doi:10.3847/1538-4357/ac78e8 , archivePrefix =. 2206.06717 , primaryClass =

work page doi:10.3847/1538-4357/ac78e8
[40]

mnras , keywords =

Super-Jeans fragmentation in massive star-forming regions revealed by triangulation analysis. mnras , keywords =. doi:10.1093/mnras/staf1116 , archivePrefix =. 2410.14610 , primaryClass =

work page doi:10.1093/mnras/staf1116
[41]

AI: Predict Density and Measure Width of molecular clouds by Multiscale Decomposition

Equation vs. AI: Predict Density and Measure Width of molecular clouds by Multiscale Decomposition. arXiv e-prints , keywords =. doi:10.48550/arXiv.2508.01130 , archivePrefix =. 2508.01130 , primaryClass =

work page doi:10.48550/arxiv.2508.01130
[42]

apj , year = 1967, month = feb, volume =

Observations of an Infrared Star in the Orion Nebula. apj , year = 1967, month = feb, volume =. doi:10.1086/149055 , adsurl =

work page doi:10.1086/149055 1967
[43]

apjl , year = 1967, month = jul, volume =

Discovery of an Infrared Nebula in Orion. apjl , year = 1967, month = jul, volume =. doi:10.1086/180039 , adsurl =

work page doi:10.1086/180039 1967
[44]

J., Moeckel, N., et al

Explosive Outflows Powered by the Decay of Non-hierarchical Multiple Systems of Massive Stars: Orion BN/KL. apj , keywords =. doi:10.1088/0004-637X/727/2/113 , archivePrefix =. 1011.5512 , primaryClass =

work page doi:10.1088/0004-637x/727/2/113
[45]

apjl , keywords =

A Tale of Three: Magnetic Fields along the Orion Integral-shaped Filament as Revealed by the JCMT BISTRO Survey. apjl , keywords =. doi:10.3847/2041-8213/ad93d2 , archivePrefix =. 2412.17716 , primaryClass =

work page doi:10.3847/2041-8213/ad93d2 2041
[46]

apjl , keywords =

On the Density Distribution in Star-forming Interstellar Clouds. apjl , keywords =. doi:10.1088/2041-8205/727/1/L20 , archivePrefix =. 1007.2950 , primaryClass =

work page doi:10.1088/2041-8205/727/1/l20 2041
[47]

mnras , keywords =

Constructing multiscale gravitational energy spectra from molecular cloud surface density PDF - interplay between turbulence and gravity. mnras , keywords =. doi:10.1093/mnras/stw1544 , archivePrefix =. 1603.04342 , primaryClass =

work page doi:10.1093/mnras/stw1544
[48]

arXiv e-prints , keywords =

Volume Density Mapper: 3D Density Reconstruction Algorithm for Molecular Clouds. arXiv e-prints , keywords =. doi:10.48550/arXiv.2509.17369 , archivePrefix =. 2509.17369 , primaryClass =

work page doi:10.48550/arxiv.2509.17369
[49]

apj , keywords =

Equation versus AI: Predicting Density and Measuring Width of Molecular Clouds by Multiscale Decomposition. apj , keywords =. doi:10.3847/1538-4357/ae3377 , archivePrefix =. 2508.01130 , primaryClass =

work page doi:10.3847/1538-4357/ae3377
[50]

VISION - Vienna survey in Orion. III. Young stellar objects in Orion A. , keywords =. doi:10.1051/0004-6361/201832577 , archivePrefix =. 1810.00878 , primaryClass =

work page doi:10.1051/0004-6361/201832577
[51]

apj , keywords =

Far-Infrared and Submillimeter Polarization of OMC-1: Evidence for Magnetically Regulated Star Formation. apj , keywords =. doi:10.1086/305139 , adsurl =

work page doi:10.1086/305139
[52]

apj , keywords =

Self-similar collapse of isothermal spheres and star formation. apj , keywords =. doi:10.1086/155274 , adsurl =

work page doi:10.1086/155274
[53]

apjl , keywords =

What Determines the Density Structure of Molecular Clouds? A Case Study of Orion B with Herschel. apjl , keywords =. doi:10.1088/2041-8205/766/2/L17 , archivePrefix =. 1304.0327 , primaryClass =

work page doi:10.1088/2041-8205/766/2/l17 2041
[54]

apj , keywords =

A New Method for Constraining Molecular Cloud Thickness: A Study of Taurus, Perseus, and Ophiuchus. apj , keywords =. doi:10.1088/0004-637X/811/1/71 , archivePrefix =. 1508.04220 , primaryClass =

work page doi:10.1088/0004-637x/811/1/71
[55]

mnras , keywords =

Magnetic, kinetic, and transition regime: spatially segregated structure of compressive MHD turbulence. mnras , keywords =. doi:10.1093/mnras/staf1320 , archivePrefix =. 2409.02769 , primaryClass =

work page doi:10.1093/mnras/staf1320
[56]

IEEE Proceedings , keywords =

Mad Max: Affine Spline Insights Into Deep Learning: In this article, the bridge between deep networks (DNs) and approximation theory via spline functions and operators is rigorously established. IEEE Proceedings , keywords =. doi:10.1109/JPROC.2020.3042100 , archivePrefix =. 1805.06576 , primaryClass =

work page doi:10.1109/jproc.2020.3042100 2020
[57]

CoRR , volume =

Learning in High Dimension Always Amounts to Extrapolation. arXiv e-prints , keywords =. doi:10.48550/arXiv.2110.09485 , archivePrefix =. 2110.09485 , primaryClass =

work page doi:10.48550/arxiv.2110.09485
[58]

apj , keywords =

Denoising Diffusion Probabilistic Models to Predict the Density of Molecular Clouds. apj , keywords =. doi:10.3847/1538-4357/accae5 , archivePrefix =. 2304.01670 , primaryClass =

work page doi:10.3847/1538-4357/accae5
[59]

Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges

Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges. arXiv e-prints , keywords =. doi:10.48550/arXiv.2104.13478 , archivePrefix =. 2104.13478 , primaryClass =

work page internal anchor Pith review doi:10.48550/arxiv.2104.13478
[60]

Toward causal representation learning.Proceedings of the IEEE, 109(5):612–634, 2021

Toward Causal Representation Learning: This article reviews fundamental concepts of causal inference and relates them to crucial open problems of machine learning, including transfer learning and generalization, thereby assaying how causality can contribute to modern machine learning research. IEEE Proceedings , keywords =. doi:10.1109/JPROC.2021.3058954 ...

work page doi:10.1109/jproc.2021.3058954 2021
[61]

Reviews of Modern Physics , keywords =

Control of star formation by supersonic turbulence. Reviews of Modern Physics , keywords =. doi:10.1103/RevModPhys.76.125 , archivePrefix =. astro-ph/0301093 , primaryClass =

work page doi:10.1103/revmodphys.76.125
[62]

mnras , keywords =

Rotation-invariant convolutional neural networks for galaxy morphology prediction. mnras , keywords =. doi:10.1093/mnras/stv632 , archivePrefix =. 1503.07077 , primaryClass =

work page doi:10.1093/mnras/stv632
[63]

mnras , keywords =

Galaxy Zoo DECaLS: Detailed visual morphology measurements from volunteers and deep learning for 314 000 galaxies. mnras , keywords =. doi:10.1093/mnras/stab2093 , archivePrefix =. 2102.08414 , primaryClass =

work page doi:10.1093/mnras/stab2093
[64]

The frontier of simulation-based inference

The frontier of simulation-based inference. Proceedings of the National Academy of Science , keywords =. doi:10.1073/pnas.1912789117 , archivePrefix =. 1911.01429 , primaryClass =

work page doi:10.1073/pnas.1912789117 1911
[65]

Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations

Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational Physics , keywords =. doi:10.1016/j.jcp.2018.10.045 , adsurl =

work page doi:10.1016/j.jcp.2018.10.045 2018
[66]

Kevrekidis, Lu Lu, Paris Perdikaris, Sifan Wang, and Liu Yang

Physics-informed machine learning. Nature Reviews Physics , year = 2021, month = jun, volume =. doi:10.1038/s42254-021-00314-5 , adsurl =

work page doi:10.1038/s42254-021-00314-5 2021
[67]

2019, arXiv preprint arXiv:1904.07248

Machine Learning in Astronomy: a practical overview. arXiv e-prints , keywords =. doi:10.48550/arXiv.1904.07248 , archivePrefix =. 1904.07248 , primaryClass =

work page doi:10.48550/arxiv.1904.07248 1904
[68]

IEEE Transactions on Knowledge and Data Engineering , keywords =

Theory-Guided Data Science: A New Paradigm for Scientific Discovery from Data. IEEE Transactions on Knowledge and Data Engineering , keywords =. doi:10.1109/TKDE.2017.2720168 , adsurl =

work page doi:10.1109/tkde.2017.2720168 2017
[69]

Camps-Valls, B

Deep learning and process understanding for data-driven Earth system science , author=. Nature , volume=. doi:10.1038/s41586-019-0912-1 , year=

work page doi:10.1038/s41586-019-0912-1
[70]

2012, in Conference on Intelligent Data Understanding (CIDU), 47 –54, doi: 10.1109/CIDU.2012.6382200

Introduction to astroML: Machine learning for astrophysics. Proceedings of Conference on Intelligent Data Understanding (CIDU , year = 2012, month = oct, pages =. doi:10.1109/CIDU.2012.6382200 , archivePrefix =. 1411.5039 , primaryClass =

work page doi:10.1109/cidu.2012.6382200 2012
[71]

2014, in Protostars and Planets VI, ed

From Filamentary Networks to Dense Cores in Molecular Clouds: Toward a New Paradigm for Star Formation. Protostars and Planets VI , year = 2014, editor =. doi:10.2458/azu_uapress_9780816531240-ch002 , archivePrefix =. 1312.6232 , primaryClass =

work page doi:10.2458/azu_uapress_9780816531240-ch002 2014
[72]

, keywords =

Alert Classification for the ALeRCE Broker System: The Real-time Stamp Classifier. , keywords =. doi:10.3847/1538-3881/ac0ef1 , archivePrefix =. 2008.03309 , primaryClass =

work page doi:10.3847/1538-3881/ac0ef1 2008
[73]

arXiv e-prints , keywords =

Shortcut Learning in Deep Neural Networks. arXiv e-prints , keywords =. doi:10.48550/arXiv.2004.07780 , archivePrefix =. 2004.07780 , primaryClass =

work page doi:10.48550/arxiv.2004.07780 2004
[74]

Kashinath, M

Physics-informed machine learning: case studies for weather and climate modelling. Philosophical Transactions of the Royal Society of London Series A , year = 2021, month = apr, volume =. doi:10.1098/rsta.2020.0093 , adsurl =

work page doi:10.1098/rsta.2020.0093 2021
[75]

apj , keywords =

The Star Formation Rate of Turbulent Magnetized Clouds: Comparing Theory, Simulations, and Observations. apj , keywords =. doi:10.1088/0004-637X/761/2/156 , archivePrefix =. 1209.2856 , primaryClass =

work page doi:10.1088/0004-637x/761/2/156
[76]

In: Advances in Neural Information Processing Systems, vol

Characterizing possible failure modes in physics-informed neural networks. arXiv e-prints , keywords =. doi:10.48550/arXiv.2109.01050 , archivePrefix =. 2109.01050 , primaryClass =

work page doi:10.48550/arxiv.2109.01050
[77]

Please stop explaining black box models for high stakes decisions.arXiv preprint arXiv:1811.10154, 2018

Stop Explaining Black Box Machine Learning Models for High Stakes Decisions and Use Interpretable Models Instead. arXiv e-prints , keywords =. doi:10.48550/arXiv.1811.10154 , archivePrefix =. 1811.10154 , primaryClass =

work page doi:10.48550/arxiv.1811.10154
[78]

Machine learning and the physical sciences,

Machine learning and the physical sciences*. Reviews of Modern Physics , keywords =. doi:10.1103/RevModPhys.91.045002 , archivePrefix =. 1903.10563 , primaryClass =

work page doi:10.1103/revmodphys.91.045002 1903
[79]

and Garcke, Jochen , journal=

Roscher, Ribana and Bohn, Bastian and Duarte, Marco F. and Garcke, Jochen , journal=. Explainable Machine Learning for Scientific Insights and Discoveries , year=
[80]

``why should i trust you?": Explaining the predictions of any classifier

``Why Should I Trust You?'': Explaining the Predictions of Any Classifier. arXiv e-prints , keywords =. doi:10.48550/arXiv.1602.04938 , archivePrefix =. 1602.04938 , primaryClass =

work page doi:10.48550/arxiv.1602.04938

Showing first 80 references.