Struct FedProxOptimizer 

pub struct FedProxOptimizer<A: Float + ScalarOperand + Debug + Send + Sync + 'static, D: Dimension + Send + Sync + Clone> { /* private fields */ }

FedProx optimizer for federated learning

FedProx adds a proximal term mu/2 * ||w - w_global||^2 to each client’s local objective, producing a gradient correction of mu * (w - w_global). This encourages local models to stay close to the global model, improving convergence under heterogeneous conditions.

Implementations

impl<A: Float + ScalarOperand + Debug + Send + Sync + 'static, D: Dimension + Send + Sync + Clone> FedProxOptimizer<A, D>

pub fn new(config: FedProxConfig<A>) -> Self

Create a new FedProxOptimizer with the given configuration

pub fn builder(num_clients: usize) -> FedProxConfigBuilder<A>

Convenience method to create a builder

pub fn set_global_parameters(&mut self, params: &[Array<A, D>]) -> Result<()>

Store the current global model parameters

This must be called before performing local updates, as the proximal term references the global parameters.

pub fn local_update( &self, params: &[Array<A, D>], gradients: &[Array<A, D>], lr: A, ) -> Result<Vec<Array<A, D>>>

Perform a local update step with the FedProx proximal gradient

Computes: w_new = w - lr * (gradient + mu * (w - w_global))

When mu=0, this reduces to standard SGD: w_new = w - lr * gradient

pub fn compute_proximal_gradient( &self, params: &[Array<A, D>], ) -> Result<Vec<Array<A, D>>>

Compute the proximal gradient: mu * (params - global_params)

This gradient term pulls local parameters towards the global model, preventing excessive divergence during local training.

pub fn submit_client_update( &mut self, client_id: usize, params: &[Array<A, D>], data_size: usize, ) -> Result<()>

Submit a client’s updated parameters after local training

pub fn aggregate_updates(&mut self) -> Result<Vec<Array<A, D>>>

Aggregate client updates using weighted averaging (FedAvg-style)

Each client’s contribution is weighted by its data_size relative to the total data across all participating clients. This produces a new global model for the next communication round.

pub fn get_round_count(&self) -> usize

Get the number of completed communication rounds

pub fn get_config(&self) -> &FedProxConfig<A>

Get a reference to the current configuration

pub fn get_global_parameters(&self) -> Option<&Vec<Array<A, D>>>

Get a reference to the current global parameters, if set

pub fn get_pending_updates_count(&self) -> usize

Get the number of client updates collected so far in this round

Trait Implementations

impl<A: Debug + Float + ScalarOperand + Debug + Send + Sync + 'static, D: Debug + Dimension + Send + Sync + Clone> Debug for FedProxOptimizer<A, D>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.