DualSparse-MoE cuts about 25% of MoE computation after training using expert partition and static neuron reconstruction, with 0.08-0.28% average accuracy loss and up to 1.41x module speedup.
How Dust Models Shape High-z Galaxy Morphology: Insights from the NewCluster Simulation
1 Pith paper cite this work. Polarity classification is still indexing.
abstract
Dust plays a pivotal role in shaping the observed morphology of galaxies. While traditional cosmological simulations often assume a fixed dust-to-gas (DTG) or dust-to-metal (DTM) mass ratio to model dust effects, recent advancements have enabled on-the-fly (OTF) dust modeling that captures the spatial and temporal evolution of dust. In this work, we investigate the impact of dust modeling on galaxy morphology using the NewCluster simulation, which implements a detailed OTF dust model. We generate mock images of NewCluster galaxies under both OTF and fixed DTM models using the radiative transfer code SKIRT, and compare their morphology to JWST observations. We measure morphology indices and use the $G-M_{20}$ test to classify galaxies. We find that the OTF galaxy models exhibit brighter centers and more pronounced bulges than those of the fixed DTM models, resulting in a lower late-type galaxy (LTG) fraction, particularly at high redshifts. This central brightening is linked to a phenomenon we refer to as the DTM cavity, a localized depression in the DTM ratio driven by intense bulge starbursts. Our results highlight the importance of modeling dust evolution in a physically motivated manner, as fixed DTM models fail to capture key morphological features.
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
DualSparse-MoE: Coordinating Tensor/Neuron-Level Sparsity with Expert Partition and Reconstruction
DualSparse-MoE cuts about 25% of MoE computation after training using expert partition and static neuron reconstruction, with 0.08-0.28% average accuracy loss and up to 1.41x module speedup.