CPLUOS Astrophysics team meeting

Friday 23 Sept 2022, 14:00 → 15:20 Asia/Seoul

물리학과

https://uos-ac-kr.zoom.us/j/8264902652

14:00 → 15:10 COSMOLOGY¶

14:00 Hyeonmo's Report - Introduction of GALIC code¶

Speaker: Hyeonmo Koo (University of Seoul)

Head-on Collision FDM/CDM Halos

• Made halo.
  • M = 2π x 10⁸ M_☉
  • r_1/2 = 0.53 kpc
  • T = 0.8 x (9.8 x 10⁸ yr / h)
• Half-mass radius evolution
  • Histogram of r for particles
    • x axis: r [kpc]
    • t = 0.0
    • t = 0.2
    • t = 0.4
    • t = 0.6
    • t = 0.8
  • Histogram of v for particles
    • x axis: v [km / s]
    • t = 0.0
    • t = 0.2
    • t = 0.4
    • t = 0.6
    • t = 0.8

14:10 Young's report. FoF and MST algorithm¶

Speaker: Young Ju

Mulguisin (MGS) Clustering Algorithm 1. Comparison of Clustering Algorithms for Study of Cosmic Structure Finding (Writing Mulguisin paper)

• Controlled data 1 - R 1.
• Controlled data 1 - R 10.
• Change 'sea level': Change minimum number of cluster (50 → 300).
• The distribution of clusters in the 3-D space.
  • MGS
  • MST
  • FoF
  • DBSCAN
• Linking length - number of cluster graph → 50 galaxies
  • This graph has vertical line. → Because of separation

14:20 Seyeon's Report - CLML¶

Speaker: Se Yeon Hwang (Universe of Seoul)

Cosmology with Large scale structure using Machine Learning (CLML)

• Data → CNN
• Data → Vit
• Started 2 Gpc.
  • 1G → x 10 times → 2G
    • Number of dark matter halos
      • 1G = 136,165
      • 2G = 1350320
• ra, dec and observed redshift like usual → Histogram bin size = (64, 64, 64)
• xyz with alcock paczynski effect → Histogram bin size = (64, 64, 64)
• w0: Good prediction
• Vit structure
• CNN structure
• Vit has almost 6 times more than CNN.
• Vit training data = 3184 (validation = 10%)
• Testing data = 796
• 6 parameters are too many.

14:30 Sumi's report¶

Speaker: Sumi Kim (University of Seoul)

Cosmology of High-Order Statistics (CHOA)

• No progress

14:40 John Suarez's report¶

Join Photo + Spect Data to Predict Physical Features

• From DESI survey
• Combine photo + spect data.
• First, create a DESI + DES test catalog.
  • Fuji SV3
    • The DESI 1% survey
• Benchmark on spectra
  • To create the dataset for DESI Fuji SV3
    • ~ 67 K (with targets)
• Input data: 65,671 objects
  • Train: 70%
  • Valid: 30%
  • Test: 30%
• Photometric normalized.
  • Epoch: 10
  • Learning rate: 0.01
  • Batch size: 32

14:50 Dr.Sabiu's Report¶

Speaker: Dr Cristiano Sabiu (University of Seoul)

Deep Learning the Deep Sky: Recovering low surface brightness objects with machine learning

• No progress

15:00 → 15:10 Discussion¶