Application of Convolutional Neural Networks to Predict Magnetic Fields’ Directions in Turbulent Clouds

We adopt the deep learning method CASI-3D (Convolutional Approach to Structure Identification-3D) infer orientation of magnetic fields in sub-/trans- Alfvenic turbulent clouds from molecular line emission. carry out magnetohydrodynamic simulations with different field strengths and use these generate synthetic observations. apply 3D radiation transfer code RADMC-3d model 12CO 13CO (J = 1-0) emission simulated then train a on data cubes predict morphology at pixel level. The trained is able directions low error (< 10deg for sub-Alfvenic samples <30deg trans-Alfvenic samples). furthermore test performance real region Taurus. prediction consistent direction inferred Planck dust polarization measurements. our developed methods produce new map Taurus that has three-times higher angular resolution than map.