Coherence Loss

Coherence Loss#

Inter-branch Coherence Loss#

Controls dispersion among K projected futures to promote stable projections.

Two modes are supported:

  • "variance" (default): mean squared deviation from the per-dimension mean.

  • "mad": mean absolute deviation from the per-dimension mean.

class qmlhc.loss.coherence.CoherenceLoss(mode='variance')[source]#

Bases: LossFn

Penalize dispersion across candidate future branches.

Parameters:

mode (str, optional) – Dispersion metric to use. Options: - "variance" (scale-sensitive, smooth) - "mad" (mean absolute deviation) Default is "variance".

Contents