EWC

class capymoa.ocl.strategy.EWC

Bases: BatchClassifier, Module, TrainTaskAware, TestTaskAware

Elastic Weight Consolidation learner.

Elastic Weight Consolidation (EWC) is a regularisation-based continual learning strategy that mitigates catastrophic forgetting by penalising changes to important parameters for previous tasks [1]. We incorporate Online EWC-style [2] updates to the Fisher diagonals, which decay the importance of previous tasks’ parameters over time based on the gamma hyperparameter.

Usually the EWC strategy has access to the entire active task’s data when estimating the Fisher diagonals, but instead we use a replay buffer to approximate the active task distribution.

[1]

Kirkpatrick, J., Pascanu, R., Rabinowitz, N., Veness, J., Desjardins, G., Rusu, A. A., Milan, K., Quan, J., Ramalho, T., Grabska-Barwinska, A., Hassabis, D., Clopath, C., Kumaran, D., & Hadsell, R. (2017). Overcoming catastrophic forgetting in neural networks. Proceedings of the National Academy of Sciences, 114(13), 3521–3526. https://doi.org/10.1073/pnas.1611835114

[2]

Schwarz, J., Czarnecki, W., Luketina, J., Grabska-Barwinska, A., Teh, Y. W., Pascanu, R., & Hadsell, R. (2018). Progress & Compress: A scalable framework for continual learning. In J. G. Dy & A. Krause (Eds.), Proceedings of the 35th International Conference on Machine Learning, ICML 2018, Stockholmsmässan, Stockholm, Sweden, July 10-15, 2018 (Vol. 80, pp. 4535–4544). PMLR. http://proceedings.mlr.press/v80/schwarz18a.html

__init__(

schema: Schema,

model: Module,

optimiser: Optimizer,

lambda_: float,

fim_buffer: int = 256,

fim_batch_size: int = 32,

device: device = torch.device('cpu'),

mask_test: bool = False,

mask_train: bool = False,

gamma: float = 1.0,

task_mask: Tensor | None = None,

) → None

Construct an EWC learner.

Parameters:

• schema – Stream schema used by the classifier interface.

• model – Torch model that outputs class logits.

• optimiser – Optimiser used to update model parameters.

• lambda – Weight of the EWC regularisation term.

• fim_buffer – Replay window size for Fisher estimation.

• fim_batch_size – Mini-batch size used when estimating Fisher diagonals.

• device – Compute device.

• mask_test – Whether to apply per-task masking during testing. This is a task incremental scenario.

• mask_train – Whether to apply per-task masking during training. This is also known as the labels trick.

• task_mask – Optional per-task mask applied to output logits.

Raises:

ValueError – If task-specific masking is requested without task_mask.

batch_predict(x: Tensor) → Tensor

Predict the labels for a batch of instances.

Parameters:

x – Batch of x_dtype valued feature vectors (batch_size, num_features)

Returns:

Predicted batch of y_dtype valued labels (batch_size,).

batch_predict_proba(x: Tensor) → Tensor

Predict the probabilities of the classes for a batch of instances.

Parameters:

x – Batch of x_dtype valued feature vectors (batch_size, num_features)

Returns:

Batch of x_dtype valued predicted probabilities (batch_size, num_classes).

batch_train(x: Tensor, y: Tensor) → None

Train with a batch of instances.

Parameters:

• x – Batch of x_dtype valued feature vectors (batch_size, num_features)

• y – Batch of y_dtype valued labels (batch_size,).

on_test_task(task_id: int) → None

Called when testing on a task starts.

on_train_task(task_id: int) → None

Called when a new training task starts.

predict(instance: Instance) → int | None

Predict the label of an instance.

The base implementation calls predict_proba() and returns the label with the highest probability.

Parameters:

instance – The instance to predict the label for.

Returns:

The predicted label or None if the classifier is unable to make a prediction.

predict_proba(

instance: Instance,

) → ndarray[tuple[Any, ...], dtype[float64]] | None

Calls batch_predict_proba() with a batch of size 1.

train(instance: LabeledInstance) → None

Calls batch_train() with a batch of size 1.

T_destination = ~T_destination

call_super_init: bool = False

device: torch.device = device(type='cpu')

Device on which the batch will be processed.

dump_patches: bool = False

random_seed: int

The random seed for reproducibility.

When implementing a classifier ensure random number generators are seeded.

schema: Schema

The schema representing the instances.

training: bool

x_dtype: torch.dtype = torch.float32

Data type for the input features.

y_dtype: torch.dtype = torch.int64

Data type for the target value/labels.