Consistency Loss

Consistency Loss#

Triadic Consistency Loss#

Enforces coherence among past, present, and projected future states (S_{t-1}, S_t, Ŝ_{t+1}).

This loss penalizes deviations between the current state and both its immediate predecessor and projected successor, promoting smooth temporal evolution in hypercausal models.

class qmlhc.loss.consistency.ConsistencyLoss(alpha=1.0, beta=1.0)[source]#

Bases: LossFn

Compute triadic consistency loss.

Penalizes incoherence between the previous, current, and projected future states. The overall loss is a weighted sum of the mean squared deviations between consecutive temporal states.

Parameters:

  • alpha (float, optional) – Weight for the present–past term, by default 1.0.

  • beta (float, optional) – Weight for the present–future term, by default 1.0.

Notes

This formulation stabilizes dynamic state transitions by encouraging local temporal consistency: L = α‖S_t - S_{t-1}‖² + β‖S_t - Ŝ_{t+1}‖².

Contents