NeurIPS paper

[Justinezgh]

Paper

NPE with score

Differentiable NPE

Smooth NF

• [x] plot of two moons learned with score matching loss

ODE Flow

• [x] plot of two moons learned with score matching loss

• [ ] test SBI with ODE Flow

Diffusion model

• [ ] plot of two moons learned with score matching loss

• [ ] test SBI with diffusion model

When do gradients help?

Constraint on the shape

• [ ] show that in the case of narrow proposal, gradients help more than with a wide proposal (Lotka Volterra)

  

• [ ] same but with more complex posterior shape -> banana shape (lensing)?

SNR impacts

• [x] #38

• [ ] understand why with large score weight we have this behavior:

  

• [ ] find a rule to say 'if SNR < value' gradients won't help

Sequential methods with score (an effective way of using score)

• [ ] #42
• [ ] propose strategy to use the score in sequential methods

Experiments

• [x] Tasks

  • [x] Galton Board
  • [x] Lotka Volterra
  • [x] Lensing (Normal)

• [x] Metrics

  • [x] c2st

• [ ] Figures Goal:

do this for each NDE?

[Justinezgh]

Code

• [ ] #41