SMLD:
score estimation
      +
Langevin Dynamics

DDPM:
forward corruption steps
      +
learned reverse probability model