iterative ref: learn white noise (sigma) iteratively

[geoffwoollard]

See Nelson eq. 8.3

3

[geoffwoollard]

Should be able to implement inside compute_bayesian_weights

• after compute bayes weights
• weight residual norm.
  • have part of residual from log_bayesian_weights = scale * (slices_norm - 2 * corr_slices_particle)
  • just need to add in norm of X (data). can precompute
  • divide through by all particles and pixels, and update at end of round. so can compute sigma_i and divide by the number of particles at the end