bound_upper_power

bound_upper_power(param: Tensor, update: Tensor, limit: float, *, power: float, **kwargs) → Tensor

Computes the scaled update of upper-bound power parameter dependence.

This is sometimes also referred to as “soft parameter dependence”.

\[U_+ = (P_\text{max} - P)^{\mu_+} U_+\]

Parameters:

• param (torch.Tensor) – parameter with update bounding, \(P\).

• update (torch.Tensor) – potentiative update being applied, \(U_+\).

• limit (float) – value of the upper bound, \(P_\text{max}\).

• power (float) – exponent of parameter dependence, \(\mu_+\).

Returns:

bounded update.

Return type:

torch.Tensor