Struct ContinualLearningSystem 

pub struct ContinualLearningSystem<T: Float + Debug + Send + Sync + 'static> { /* private fields */ }

Continual learning system backed by crate::continual_learning::ElasticWeightConsolidation.

Implementations

impl<T: Float + Debug + Send + Sync + 'static + Default + Clone + Sum + ScalarOperand> ContinualLearningSystem<T>

pub fn new(settings: &ContinualLearningSettings) -> Result<Self>

pub fn with_steps_per_task(self, steps: usize) -> Result<Self>

Override the number of gradient steps taken per task.

pub fn learn_sequence( &mut self, sequence: &[MetaTask<T>], meta_parameters: &mut HashMap<String, Array1<T>>, ) -> Result<ContinualLearningResult<T>>

Learn a sequence of tasks with elastic weight consolidation.

Delegates to crate::continual_learning::ElasticWeightConsolidation: each task is trained with SGD on the shared linear model plus the EWC penalty gradient, then consolidated (Fisher diagonal + anchor). Forgetting is measured for real by re-evaluating every earlier task after the sequence finishes.

pub fn forgetting_measure(&self) -> T

Forgetting measured by the most recent Self::learn_sequence call.

Zero before any sequence has been learned.