Generative Models in Astrophysics
Carol Cuesta-Lazaro
IAIFI Fellow
Initial Conditions of the Universe
Laws of gravity
3-D distribution of galaxies
Which are the ICs of OUR Universe?
Primordial non-Gaussianity?
Probe Inflation
Galaxy formation
3-D distribution of dark matter
Is GR modified on large scales?
How do galaxies form?
Neutrino mass hierarchy?
Dall-e 3
A 2D animation of a folk music band composed of anthropomorphic autumn leaves, each playing traditional bluegrass instruments, amidst a rustic forest setting dappled with the soft light of a harvest moon
Emulate complex processes in high dimensions
x∼p(x∣y)
x \sim p(x|y)
y={Ωcdm,σ8,...}
y = \{ \Omega_{cdm}, \sigma_8, ... \}
arXiv:2206.04594
Generative AI for physics
Super resolution
arxiv:2010.06608
Generative AI for physics
Compare data and simulations optimally
Simulation-based Inference
arXiv:1911.01429
Generative AI for physics
Detect anomalies
arxiv:2010.14554
Generative AI for physics
Model uncertainties
arxiv:2010.14554
Current Universe
Initial Conditions
Generative AI for physics
Represent complex priors
arxiv:2206.14820
how does a galaxy look like?
lensing
Generative AI for physics
pϕ(x)
p_\phi(x)
Data
A parametric PDF
Maximise the likelihood
(or something similar)
Generative AI for physics
2. Sample
x∼pϕ(x)
x \sim p_\phi(x)
1. Estimate densities
pϕ(x)
p_\phi(x)
Explicit Density
Implicit Density
Tractable Density
Approximate Density
Normalising flows
Variational Autoencoders
Diffusion models
Generative Adversarial Networks
The zoo of generative models
The backbone of vision generative models
Reverse diffusion: Denoise previous step
Forward diffusion: Add Gaussian noise (fixed)
A person half Yoda half Gandalf
Diffusion Models
zT
z_T
z0
z_{0}
z1
z_{1}
Diffusion on point clouds
z2
z_{2}
qθ(zt−1∣zt)
q_\theta(z_{t-1}|z_t)
p(zt∣zt−1)
p(z_t|z_{t-1})
Reverse diffusion: Denoise previous step
Forward diffusion: Add Gaussian noise (fixed)
Cosmology
"Diffusion generative modeling for galaxy surveys: emulating clustering for inference at the field level" Carolina Cuesta-Lazaro, Siddarth Mishra-Sharma
Generating galaxy point clouds for redshift surveys
Setting tight constraints with only 5000 halo positions
p(x,y,z,vx,vy,vz,Mh∣Ωm,σ8)
p(x,y,z, v_x, v_y, v_z, M_h|\Omega_m, \sigma_8)
Halo Mass Function
Velocity
Mean pairwise velocity
(Probabilistic) reconstruction of Dark Matter
"Probabilistic Reconstruction of Dark Matter fields from galaxies using diffusion models" Victoria Ono, Core Francisco Park, Nayantara Mudur, Yueying Ni, Carolina Cuesta-Lazaro (in prep)
pϕ(xDM∣xStars)
p_\phi(x_\mathrm{DM}|x_\mathrm{Stars})
(Probabilistic) reconstruction of Dark Matter
(Probabilistic) reconstruction of Dark Matter
(Probabilistic) reconstruction of Dark Matter
"Probabilistic Reconstruction of Dark Matter fields from galaxies using diffusion models" Victoria Ono, Core Francisco Park, Nayantara Mudur, Yueying Ni, Carolina Cuesta-Lazaro (in prep)
Differentiable PM Simulators
h0
h_0
h1
h_1
h5
h_5
h4
h_4
h2
h_2
h3
h_3
h6
h_6
Node features coordinates (+mass, velocities)
Input
Noisy halo properties
Output
Noise prediction
Graph Neural Networks as score models
kNN (~20)
Generative Models in Astrophysics Carol Cuesta-Lazaro IAIFI Fellow
deck
By carol cuesta
deck
- 361
