NullRegularizer#

class fl_sim.regularizers.NullRegularizer(coeff: float = 1.0)[source]#

Bases: Regularizer

Null regularizer, or equivalently the zero function.

Parameters:

coeff (float, default 1.0) – The coefficient of the regularizer.

eval(params: Iterable[Parameter], coeff: float | None = None) → float[source]#

Evaluate the regularizer on the given parameters.

Parameters:

  • params (Iterable[torch.nn.parameter.Parameter]) – The parameters to be evaluated on.

  • coeff (float, optional) – The coefficient of the regularizer. If None, use the default value.

prox_eval(params: Iterable[Parameter], coeff: float | None = None) → Iterable[Parameter][source]#

Evaluate the proximity operator of the regularizer on the given parameters.

i.e. evaluate the following function:

\[\mathrm{prox}_{\lambda R}(\mathbf{w}) = \arg\min_{\mathbf{u}} \frac{1}{2s} \lVert \mathbf{u} - \mathbf{w} \rVert_2^2 + \lambda R(\mathbf{u})\]

where \(R\) is the regularizer.

Parameters:

  • params (Iterable[torch.nn.parameter.Parameter]) – The parameters to be evaluated on.

  • coeff (float, optional) – The coefficient of the regularizer. If None, use the default value.

Returns:

The proximity operator of the regularizer evaluated on the given parameters.

Return type:

Iterable[torch.nn.parameter.Parameter]