CPLUOS Astrophysics team meeting

Friday 1 Apr 2022, 14:00 → 15:10 Asia/Seoul

물리학과

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

14:00 → 15:00 COSMOLOGY

14:00 Hyeonmo's Report - Head-on Collsion of FDM/CDM Halos(POSTER)

Speaker: Hyeonmo Koo (University of Seoul)

Head-on Collsion of FDM/CDM Halos(POSTER)

• lnΛ → That is not precise.

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

Speaker: Young Ju

FoF and MST algorithm

• Mulguisin Clustering Algorithm I. Comparison of Clustering Algorithms for Study of Cosmic Structure Finding
• We need Cluster finding algorithm.
• Cluster finding algorithms
  • What is the differences of FoF, MST, and DBSCAN?
    • Traditional algorithm : FoF(Friends of Friends)
    • Graph based algorithms : MST(Minimum spanning tree)
    • Machine learning(ML) algorithm : DBSCAN(Density-based spatial clustering of applications with noise)
• We propose new cluster finding algorithm, MulGuiSin(MGS)
• Schematic idea of MGS
  • 1. MGS with different height hide into the water tank.
  • 2. We drain water from the water tank.
  • 3. The highest MGS first appears.
  • 4. As draining consecutively, the next highest MGS appears.
  • 4. Keep draining until there is no remained MGS
  • → The MGS trace galaxies top-bottom approach.
• MGS algorithm
  • 1. Find galaxy with the highest density.
  • 2. The galaxy becomes a new cluster, "MGS".
  • 3. Find the next galaxy with second highest density and estimate the distance between "MGS" and next galaxy.
  • 4. If the distance is less than out criteria, the second galaxy is merged with "MGS".
  • 5. This process continues until there is no galaxy.
• Features of MGS algorithm
  • 1. MGS can find structure from galaxy data in detail.
  • 2. Provide topological information
    • Number of nodes
    • Number of branches
    • Number of children
    • Link length
    • Average node generation
    • ...
• Performance test
  • Controlled data 1 : Simple gaussian model
    • Limit of Linking-length : 4
      • Because of the meger of data
  • Controlled data 2 : Power-law model
  • Controlled data 3 : Halo mass function + NFW profile
• Conclusion
  • We propose new cluster finding algorithm MGS.
  • The MGS shows good performance in the test data set.
  • MGS can provide lots of topological information, it  can give us other perspective of research for large-scale structure.

14:20 Hannah's Report

See the slides from 8p

Speaker: Hannah Jhee (University of Seoul)

Tracking Halo Orbits and Their Mass Evolution around the Large-scale Filaments

14:30 CHOA(Cosmology of High-Order Statistics)

Speaker: Sumi Kim (University of Seoul)

CHOA(Cosmology of High-Order Statistics)

• Pinocchio halo catalogue data → 3 - point correlation function(input) → Relatively Simple Machine Learning → Ω_m, σ_8(output)
• Decision Tree
  • The goal is to sperate the classes using data's features.
  • In each node input data is seperated into N different subnodes.
  • Depth of tree = Number of questions
• Random Forest
  • Set of many individual decision trees.
  • Class with most votes becomes prediction.
  • Large uncorrelated model performs better than individual trees.
  • Gets sample of data but the same size.
  • Origianl = [1, 2, 3, 4, 5, 6]
  • Sample = [1, 2, 3, 3, 5, 5]
• 3pcf of different cosmologies
  • Ω_m = 0.16, σ_8 = 0.40(output)
  • Ω_m = 0.26, σ_8 = 0.40(output)
  • Ω_m = 0.26, σ_8 = 1.10(output)
• Test in 'different cosmology' data
  • Random Forest result is slightly better in constraint.
• Mid - term inspection in 'single cosmology' data
  • 3 - point correlation function & Deep learning using mass and mass weighted velocity
  • → It's not fair comparison due to the scale. We expect deep learning will win if using full size.
• For Next
  • Planning to stick 3pcf + 2pcf togother and use it as input.
  • Can compare with this result.

14:40 CLML

Cosmology with Large scale structure using Machine Learning

Speaker: Se Yeon Hwang

14:50 Dr.Sabiu's Report - N-PCF

Speaker: Dr Cristiano Sabiu (University of Seoul)

15:00 → 15:10 Discussion