Characterising lognormal fractional-Brownian-motion density fields with a Convolutional Neural Network

Abstract

In attempting to quantify statistically the density structure of the interstellar medium, astronomers have considered a variety of fractal models. Here we argue that, to properly characterise a fractal model, one needs to define precisely the algorithm used to generate the density field, and to specify -- at least -- three parameters: one parameter constrains the spatial structure of the field; one parameter constrains the density contrast between structures on different scales; and one parameter constrains the dynamic range of spatial scales over which self-similarity is expected (either due to physical considerations, or due to the limitations of the observational or numerical technique generating the input data). A realistic fractal field must also be noisy and non-periodic. We illustrate this with the exponentiated fractional Brownian motion (xfBm) algorithm, which is popular because it delivers an approximately lognormal density field, and for which the three parameters are, respectively, the power spectrum exponent, $\beta$, the exponentiating factor, $\mathcal{S}$, and the dynamic range, $\mathcal{R}$. We then explore and compare two approaches that might be used to estimate these parameters: Machine Learning and the established $\Delta$-Variance procedure. We show that for $2\le \beta \le 4$ and $0\le \mathcal{S}\le 3$, a suitably trained Convolutional Neural Network is able to estimate objectively both $\beta$ (with root-mean-square error ${\epsilon }_{{}_{\beta }}\sim 0.12$) and $\mathcal{S}$ (with ${\epsilon }_{{}_{\mathcal{S}}}\sim 0.29$). $\Delta$-variance is also able to estimate $\beta$, albeit with a somewhat larger error (${\epsilon }_{{}_{\beta }}\sim 0.17$) and with some human intervention, but is not able to estimate $\mathcal{S}$.

References in corpus (30)

• An overview of gradient descent optimization algorithms
• Comparing the statistics of interstellar turbulence in simulations and observations: Solenoidal versus compressive turbulence forcing
• Rotation-invariant convolutional neural networks for galaxy morphology prediction
• Cluster-formation in the Rosette molecular cloud at the junctions of filaments
• On the universality of supersonic turbulence
• What determines the density structure of molecular clouds ? A case study of Orion B with Herschel
• The Fractal Density Structure in Supersonic Isothermal Turbulence: Solenoidal versus Compressive Energy Injection
• On the evolution of the density pdf in strongly self-gravitating systems
• The Mass-Size Relation from Clouds to Cores. II. Solar Neighborhood Clouds
• Unfolding the Laws of Star Formation: The Density Distribution of Molecular Clouds
• A Fractal Origin for the Mass Spectrum of Interstellar Clouds: II. Cloud Models and Power Law Slopes
• A cluster in the making: ALMA reveals the initial conditions for high-mass cluster formation
• The Fractal Dimension of Projected Clouds
• A supersonic turbulence origin of Larson's laws
• Self-gravity as an explanation of the fractal structure of the interstellar medium
• Multifractal Scaling, Geometrical Diversity, and Hierarchical Structure in the Cool Interstellar Medium
• Structure analysis of interstellar clouds: I. Improving the Delta-variance method
• Deep Convolutional Denoising of Low-Light Images
• Mach number study of supersonic turbulence: The properties of the density field
• Structure analysis of interstellar clouds: II. Applying the Delta-variance method to interstellar turbulence
• Multifractal Analysis of the Interstellar Medium. First application to Hi-GAL Observations
• Deep Galaxy: Classification of Galaxies based on Deep Convolutional Neural Networks
• The relation between the turbulent Mach number and observed fractal dimensions of turbulent clouds
• Characterizing the structure of diffuse emission in Hi-GAL maps
• Deriving star cluster parameters with convolutional neural networks. I. Age, mass, and size
• Mass Functions in Fractal Clouds: The Role of Cloud Structure in the Stellar Initial Mass Function
• The relation between the true and observed fractal dimensions of turbulent clouds
• Single-epoch supernova classification with deep convolutional neural networks
• Modelling the structure of star clusters with fractional Brownian motion
• Particle identification in ground-based gamma-ray astronomy using convolutional neural networks

Cited by in corpus (2)

• Cloud structure and young star distribution in the Dragonfish complex
• Direct power spectral density estimation from structure functions without Fourier transforms