HAT

HAT is a parameter-isolation-based continual learning method which learns real-valued masks, which play a role similar to attention modules, for weighting each layer output for each task. For the details, see the original paper.

Node-level Problems

class NCTaskILHATTrainer(model, scenario, optimizer_fn, loss_fn, device, **kwargs)[source]
afterInference(results, model, optimizer, _curr_batch, training_states)[source]

The event function to execute some processes right after the inference step (for training). We recommend performing backpropagation in this event function.

For HAT, we need to consider the techniques introduced in the original paper in this event function.

Parameters:

  • results (dict) – the returned dictionary from the event function inference.

  • model (torch.nn.Module) – the current trained model.

  • optimizer (torch.optim.Optimizer) – the current optimizer function.

  • curr_batch (object) – the data (or minibatch) for the current iteration.

  • curr_training_states (dict) – the dictionary containing the current training states.

  • use_mask (bool) – whether model masks weights of the model.

Returns:

A dictionary containing the information from the results.

criterion(outputs, targets, masks)[source]
get_mask(task_id, model, s, m, e)[source]
get_view_for(name, parameter, masks)[source]
inference(model, _curr_batch, training_states)[source]

The event function to execute inference step.

For task-IL, we need to additionally consider task information for the inference step. In addition, we need to consider (attentive) masks for HAT.

Parameters:

  • model (torch.nn.Module) – the current trained model.

  • curr_batch (object) – the data (or minibatch) for the current iteration.

  • curr_training_states (dict) – the dictionary containing the current training states.

Returns:

A dictionary containing the inference results, such as prediction result and loss.

initTrainingStates(scenario, model, optimizer)[source]

The event function to initialize the dictionary for storing training states (i.e., intermedeiate results).

Parameters:
Returns:

Initialized training state (dict).

processAfterTraining(task_id, curr_dataset, curr_model, curr_optimizer, curr_training_states)[source]

The event function to execute some processes after training the current task.

In this event function, our implementation updates the masks for running HAT.

Parameters:

  • task_id (int) – the index of the current task.

  • curr_dataset (object) – The dataset for the current task.

  • curr_model (torch.nn.Module) – the current trained model.

  • curr_optimizer (torch.optim.Optimizer) – the current optimizer function.

  • curr_training_states (dict) – the dictionary containing the current training states.

processBeforeTraining(task_id, curr_dataset, curr_model, curr_optimizer, curr_training_states)[source]

The event function to execute some processes before training.

Parameters:

  • task_id (int) – the index of the current task

  • curr_dataset (object) – The dataset for the current task.

  • curr_model (torch.nn.Module) – the current trained model.

  • curr_optimizer (torch.optim.Optimizer) – the current optimizer function.

  • curr_training_states (dict) – the dictionary containing the current training states.

processEvalIteration(model, _curr_batch)[source]

The event function to handle every evaluation iteration.

We need to extend the base function since the output format is slightly different from the base trainer.

Parameters:

  • model (torch.nn.Module) – the current trained model.

  • curr_batch (object) – the data (or minibatch) for the current iteration.

  • use_mask (bool) – whether model masks weights of the model.

Returns:

A dictionary containing the outcomes (stats) during the evaluation iteration.

processTrainIteration(model, optimizer, _curr_batch, training_states)[source]

The event function to handle every training iteration.

Parameters:

  • model (torch.nn.Module) – the current trained model.

  • optimizer (torch.optim.Optimizer) – the current optimizer function.

  • curr_batch (object) – the data (or minibatch) for the current iteration.

  • curr_training_states (dict) – the dictionary containing the current training states.

  • use_mask (bool) – whether model masks weights of the model.

Returns:

A dictionary containing the outcomes (stats) during the training iteration.