CPLUOS Astrophysics team meeting

Friday 20 May 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

• No progress

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

Speaker: Young Ju

FoF and MST algorithm

• No progress

14:20 Hannah's Report

See the slides from 8p

Speaker: Hannah Jhee (University of Seoul)

Nyx: A Massively Parallel AMR Code for Computational Cosmology

• AxioNyx
  • Cold Dark Matter, Baryon → axio- → Cold Dark Matter, Baryon + FDM
• Install
  • Environment and prerequi8i:e8
    • opeammai-4.1.2/
    • pp:w-8.8.10/
    • amre::4a::ioay::-1.0/
• Test Run & plot with yt package

14:30 Sumi's report

Speaker: Sumi Kim (University of Seoul)

DESI  Early Data BAO Detection Project

• No progress

14:40 CLML

Cosmology with Large scale structure using Machine Learning

Speaker: Se Yeon Hwang

Cosmology with Large scale structure using Machine Learning(CLML)

• Outline
  • We're gointo show the comparison between cola and Pinocchio
  • For the theory model, we used COLOSSUS
  • Cosmological parameter (arXiv: 1712.04512v3)
    • Table 1, Pre-set cosmologies
      • ID = planck18
      • rel = yes
      • H_0 = 67.66
      • Ω_m = 0.3111
      • Ω_b = 0.0490
      • n_s = 0.9665
      • σ_8 = 0.8102
      • Reference = Planck Coll. 2018. Table 2
      • Comment = Best fit, with BAO (column 6)
  • Box size = 1 Gpc, number of particles = 512^3, minimum particles of halo = 10
  • Both are about dark matter halos.
  • This is the mass histogram that we got.
  • As you can see Pinocchio's halo is about as twice as many as cola's halo.
    • Mass histogram
    • log 10 (mass) - the number graph
    • The number of halo
      • pinocchio = 657,578
      • cola = 365,382
  • Mass function
    • Pinocchio is closer to theory than cola.
  • Power spectrum
    • Pinocchio is closer to theory than cola.
• Conclusion
  • We slightly convert our plan.
  • We was planning to use 3 different simulation (cola, Pinocchio, NECOLA (machine learning algorithm).
  • But We decided to use only PINOCCHIO.
  • Using different option in pinpcchio (previous we only used snapshot, but we can also apply with light cone).
  • Using different algorithm (CNN and ViT) would be the future plan for CLML.

14:50 Dr.Sabiu's Report

Speaker: Dr Cristiano Sabiu (University of Seoul)

Deep Learning the Deep Sky

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

• New candidates ...
• Next:
  • We are planning to update the catalogue of low mass dwarf galaxies to create the most complete and comprehensive survey of the local universe
  • A dataset that will be useful not only for galaxy physics but maybe also cosmology
• Testing Data Result
• Detailed inspection
• label: Predict class % (truth class)
• Test Data Confusion Matrix
  • Predictions - Actuals graph
• Locating the dwarf in the image
  • We can look into the network at any point.
  • Choosing one of the last convolution layers shows where in the image is highly activated.
  • We feed in an image to the trained network then look at the activation after the 4th convolution layer which has (8x8) spatial dimension.
    • Find weighted centre
    • Adjust to input image dimensions, overlay original
• Self-Attention
  • This paper revolutionised the field of natural language processing (NLP) e.g. GPT3 as we showed many months ago can write code etc.
  • The transformer network introduced in this paper is now used in many cutting edge ML models not only in NLP but also computer vision (CV).
    • Reference - Attention Is All You Need
• Vision Transformers
  • Building upon the powerful self attention transformers this paper shows a successful application to computer vision tasks like classification.
  • Cutting edge only a few years old
  • Outperforming the older convolutional neural network (CNN) models.
    • Reference - An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

15:00 → 15:10 Discussion