SearcharxivSearch

arXiv subjects

Chou

Publications and source records attributed to Chou.

2 recordsLinked to original sources

Probing the Physical and Chemical Characteristics of an Extremely Early Class 0 Protostar in G204.4-11.3A2-NE

We have observed the low-mass molecular cloud core G204.4-11.3A2-NE (G204NE) in the direction of Orion B giant molecular cloud with the Atacama Large Millimeter/submillimeter Array in Band 6. The $\rm 1.3\ mm$ continuum images and visibilities unveil a compact central structure with a radius of $\sim$12 au, while showing no signature of binarity down to 18 au. The bolometric temperature and luminosity of this source are derived to be ${\sim}$33 K and $\sim$1.15 $~L_{\odot}$, respectively. Chemical stratification is observed in dense gas tracers, with C$^{18}$O emission peaking at the continuum position surrounded by the spatially extended emission of N$_2$D$^+$ and DCO$^+$. This implies that the core is in a very early evolutionary stage in which $\rm CO$ depletion occurs in most regions except for a small area heated by the central source. The envelope kinematics indicates a rotating and infalling structure with a central protostar mass of 0.08-0.1 $M_{\odot}$. The protostar drives a collimated outflow traced by CO, SiO, SO, and H$_2$CO, with misaligned blueshifted and redshifted lobes exhibiting a pair of bow-like patterns. High-velocity jets, extending up to 720 au, are detected in CO, SiO, and SO lines. The jet launching region is likely within twice of the dust sublimation zone. The absence of a binary signature suggests the outflows and jets are driven by a single protostar, although a close binary cannot be ruled out. The observed deflection of the outflows and jet is likely due to turbulent accretion in a moderately magnetized core.

astro-ph.SR

Pre-Quantized Deep Learning Models Codified in ONNX to Enable Hardware/Software Co-Design

This paper presents a methodology to separate the quantization process from the hardware-specific model compilation stage via a pre-quantized deep learning model description in standard ONNX format. Separating the quantization process from the model compilation stage enables independent development. The methodology is expressive to convey hardware-specific operations and to embed key quantization parameters into a ONNX model which enables hardware/software co-design. Detailed examples are given for both MLP and CNN based networks, which can be extended to other networks in a straightforward fashion.

cs.LG