Multitask Regression/Classification Network Quantum State Reconstruction

Multitask regression/classification network-based quantum state reconstruction as proposed by Luu et al. (2024) [3].

qsttoolkit.tomography.dlqst.multitask_reconstructor.architecture.build_classification_tail(input_features: Tensor, num_classes: int) → Tensor

Builds the classification tail of the multitask reconstructor model.

Parameters:

• input_features (tf.Tensor) – Input feature vector.

• num_classes (int) – Number of classes to classify into.

Returns:

Output tensor of the classification tail.

Return type:

tf.Tensor

qsttoolkit.tomography.dlqst.multitask_reconstructor.architecture.build_feature_extractor(input_shape: tuple) → Model

Builds the feature extractor branch of the multitask reconstructor model.

Parameters:

input_shape (tuple) – Shape of the input data.

Returns:

Feature extractor model.

Return type:

tf.keras.Model

qsttoolkit.tomography.dlqst.multitask_reconstructor.architecture.build_multitask_reconstructor(input_shape: tuple, num_classes: int, num_regression_outputs: int) → Model

Builds the multitask reconstructor model.

Parameters:

• input_shape (tuple) – Shape of the input data.

• num_classes (int) – Number of classes to classify.

• num_regression_outputs (int) – Number of regression outputs.

Returns:

Multitask reconstructor model.

Return type:

tf.keras.Model

qsttoolkit.tomography.dlqst.multitask_reconstructor.architecture.build_regression_tail(input_features: Tensor, num_outputs: int) → Tensor

Builds the regression tail of the multitask reconstructor model.

Parameters:

• input_features (tf.Tensor) – Input feature vector.

• num_outputs (int) – Number of regression outputs.

Returns:

Output tensor of the regression tail.

Return type:

tf.Tensor

class qsttoolkit.tomography.dlqst.multitask_reconstructor.model.MultitaskQuantumStateTomography(X_train: ndarray, X_test: ndarray, y_train: ndarray, y_test: ndarray, label_encoder: LabelEncoder, early_stopping_patience: int = 30, lr_scheduler_factor: float = 0.5, lr_scheduler_patience: int = 15, dim=None)

Bases: object

A class for training and evaluating a multitask regression/classification model for quantum state tomography.

X_train

Training data.

Type:

np.ndarray

X_test

Test data.

Type:

np.ndarray

y_train

Dictionary containing both the training classification and regression labels.

Type:

dict

y_test

Dictionary containing both the test classification and regression labels.

Type:

dict

label_encoder

Label encoder for the classification labels.

Type:

sklearn.preprocessing.LabelEncoder

early_stopping_patience

Number of epochs with no improvement after which training will be stopped. Defaults to 30.

Type:

int

lr_scheduler_factor

Factor by which the learning rate will be reduced. Must be < 1. Defaults to 0.5.

Type:

float

lr_scheduler_patience

Number of epochs with no improvement after which learning rate will be reduced. Defaults to 15.

Type:

int

evaluate_classification(include_confusion_matrix: bool = True, include_classification_report: bool = True)

Evaluates the classification performance of the model.

Parameters:

• include_confusion_matrix (bool) – Whether to include the confusion matrix in the evaluation. Defaults to True.

• include_classification_report (bool) – Whether to include the classification report in the evaluation. Defaults to True.

evaluate_regression()

Evaluates the regression performance of the model by plotting the predictions vs true values for each class.

infer(data: ndarray)

Infers the quantum state label and key parameter from input data.

Parameters:

data (np.ndarray) – Input data.

Returns:

Predicted quantum state labels.

Return type:

np.ndarray

plot_training()

Plots the training history over epochs.

train(optimizer=<keras.src.optimizers.adam.Adam object>, classification_loss='sparse_categorical_crossentropy', regression_loss='mse', classification_loss_weight: float = 1.0, regression_loss_weight: float = 0.5, classification_metric: str = 'accuracy', regression_metric: str = 'mae', epochs: int = 30, batch_size: int = 32, validation_split: float = 0.2, verbose='auto')

Compiles and trains the model.

Parameters:

• optimizer – Optimizer to use in the training. Defaults to Adam with learning rate 0.0005, beta_1=0.9, and beta_2=0.999.

• classification_loss – Classification loss function to use in the training. Defaults to ‘sparse_categorical_crossentropy’.

• regression_loss – Regression loss function to use in the training. Defaults to ‘mse’.

• classification_loss_weight (float) – Weight of the classification loss in the total loss. Defaults to 1.0.

• regression_loss_weight (float) – Weight of the regression loss in the total loss. Defaults to 1.0.

• classification_metric (str) – Metric to measure classification performance during training. Defaults to ‘accuracy’.

• regression_metric (str) – Metric to measure regression performance during training. Defaults to ‘mse’.

• epochs (int) – Number of epochs to train the model. Defaults to 30.

• batch_size (int) – Training batch size. Defaults to 32.

• validation_split (float) – Fraction of the training data to use as validation data. Defaults to 0.2.

• verbose – Verbosity mode. Defaults to ‘auto’.

class qsttoolkit.tomography.dlqst.multitask_reconstructor.reconstruction.StateReconstructor

Bases: object

Class to reconstruct states from the predicted labels and key parameters. Reconstructed states are stored in the predictions_df DataFrame, along with the true states and density matrices.

add_data(true_labels: list[str], predicted_labels: list[str], true_state_parameters: list[float], predicted_state_parameters: list[float], true_dms: list[ndarray] = None, true_states=None)

Supplies the true and predicted labels and state parameters, and true state density matrices, to the predictions_df DataFrame.

Parameters:

• true_labels (list of str) – List of true labels.

• predicted_labels (list of str) – List of predicted labels.

• true_state_parameters (list of float) – List of true state parameters.

• predicted_state_parameters (list of float) – List of predicted state parameters.

• true_dms (list of np.ndarray) – List of true state density matrices.

calculate_fidelities()

Calculates the fidelities between the true and reconstructed states, and stores them in the self.predictions_df DataFrame ‘fidelity’ column.

plot_Husimi_Qs(state_range: list[int, int], xgrid: ndarray = None, ygrid: ndarray = None)

Deprecated alias for plot_comparison_Husimi_Qs.

plot_comparison_Hintons(state_range: list[int, int])

Plots Hinton diagrams of the true and reconstructed density matrices for a given range of states.

Parameters:

state_range (list of int) – List of two integers, the minimum and maximum state indices to plot.

plot_comparison_Husimi_Qs(state_range: list[int, int], xgrid: ndarray = None, ygrid: ndarray = None)

Plots Husimi Q functions of the true and reconstructed states for a given range of states.

Parameters:

• state_range (list of int) – List of two integers, the minimum and maximum state indices to plot.

• xgrid (np.ndarray) – Grid for the real part of the coherent state parameter. Defaults to np.linspace(-5, 5, 100).

• ygrid (np.ndarray) – Grid for the imaginary part of the coherent state parameter. Defaults to np.linspace(-5, 5, 100).

plot_fidelities(color_by_true_label: bool = False)

Plots a histogram of the fidelities between the true and reconstructed states.

Parameters:

color_by_true_label (bool) – If True, the fidelities are colored by the true label. Defaults to False.

plot_hintons(state_range: list[int, int])

Deprecated alias for plot_comparison_Hintons.

reconstruct(Nc=None)

Reconstructs the states from the restricted predicted state parameters, and stores the reconstructed states and density matrices in the self.predictions_df DataFrame.

restrict_parameters(fock_n_range: list[int, int], binomial_S_range: list[int, int])

Restricts the predicted state parameters to be within a certain set range, depending on the predicted label, in order to enforce physicality of reconstructed states. Restricted predicted state parameters are stored in the self.predictions_df DataFrame.

Parameters:

• fock_n_range (list of int) – List of two integers, the minimum and maximum Fock state parameter values.

• binomial_S_range (list of int) – List of two integers, the minimum and maximum binomial state parameter values.