QUBIC

Minutes:

• New machine learning approach for noise (Noah):
  - NODE (Neural Ordinary Differential Equation) with an added relaxation time (see Ornstein-Uhlenbeck in statistal physics)
  --> in order not to depend on the initial conditions of the noise generation, this NN has a relaxation time that makes it converge to the intended noise

• convolution noise (Alexandre):
  - hard to run jobs on CC, a lot of demand
  - nside=1024 is too large for the CC, might need 1TB of memory
  - maybe try nside=512? See if it's ok in terms of fwhm_mapmaking, and try on CC with 200-300GB?
  - see if memory used is proportinal to nside by comparing nside 64, 124, 256
  - maybe adjust the ntasks-per-node, cpus-per-node, time...