CLML(Cosmology with Large scale structure using Machine Learning)
- Outlook
- Started on the machine learning!(arXIv : 1908.10590) which is using 3d convolution network
- input(3d density field of dark matter halo) → CNN CNN CNN Fully connected Fully connected Fully connected → output(Ω_m, σ_8 → Cosmological parameter)
- Data
- xyz position of dark matter halo
- Box size = 256(Mpc/h)
- Number of particles = 256^3
- 3d histogram of 32 voxels(density field)
- 1 voxel have 8(Mpc/h) information
- Change for the data
- 128
- Boxsize = 256(Mpc/h)
- Number of particles = 128^3
- 256
- Boxsize = 256(Mpc/h)
- Number of particles = 256^3
- 512
- Boxsize = 256(Mpc/h)
- Number of particles = 512^3
- Data
- Number of halo
- The number of halo(128) : 9225
- The number of halo(256) : 74646
- The number of halo(512) : 504649
- Halo mass
- Ω_m = 0.33, σ_8 = 0.62
- ρ = N(↑)*M(↓)/V
- 1 voxel have 2(Mpc/h) information using 32 voxel(then one side should be 64(Mpc/h)
- Short version
- Only using 0(Mpc/h) < x, y, z < 64 (Mpc/h)
- Including only 0~64 information
- Cut version
- At each side : [0, 64] [64, 128] [128, 192] [192 256] >> 32 voxel histogram
- Including all information of the box scale
- Total 64(4^3) for one cosmology
- Data set
- 128_short, 256_short, 512_short, 128_cut, 256_cut, 512_cut
- Model
- Model 1
- Model 2
- Model 3
- Model 4
- Conclusion
- Don't sure about what is going on.
- There are possibility of my mistakes + possibility of different results each time when run it even
- Used same model and same data(I have to check)
- Next step
- Wil try running with different cosmology + same seed
- Also, have to understand about mathematics on how each layer works
