MAP Landscape Tractability

Background: For non-impulsive, irreversible processes, the numerical optimization landscape for finding the Maximum A Posteriori (MAP) estimate can become challenging due to competing local minima.

Question / Future Work: For some irreversible cases, the MAP optimization landscape becomes rugged, characterized by high multi-start dispersion, preventing full exploitation of the temporal asymmetry signal. Future work should focus on improving MAP landscape tractability, possibly by adopting normalizing flow backbones whose bijective structure offers a smoother optimization space, or by using score-based diffusion models.