Rede neural convolucional quântica

Veja no TensorFlow.org Executar no Google Colab Ver fonte no GitHub Baixar caderno

Este tutorial implementa uma Rede Neural Convolucional Quântica (QCNN) simplificada, um análogo quântico proposto para uma rede neural convolucional clássica que também é translacionalmente invariável .

Este exemplo demonstra como detectar certas propriedades de uma fonte de dados quântica, como um sensor quântico ou uma simulação complexa de um dispositivo. A fonte de dados quântica sendo um estado de cluster que pode ou não ter uma excitação - o que o QCNN aprenderá a detectar (o conjunto de dados usado no artigo foi a classificação de fase SPT).

Configurar

pip install tensorflow==2.7.0

Instale o TensorFlow Quantum:

pip install tensorflow-quantum
# Update package resources to account for version changes.
import importlib, pkg_resources
importlib.reload(pkg_resources)
<module 'pkg_resources' from '/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/pkg_resources/__init__.py'>

Agora importe o TensorFlow e as dependências do módulo:

import tensorflow as tf
import tensorflow_quantum as tfq

import cirq
import sympy
import numpy as np

# visualization tools
%matplotlib inline
import matplotlib.pyplot as plt
from cirq.contrib.svg import SVGCircuit
2022-02-04 12:43:45.380301: E tensorflow/stream_executor/cuda/cuda_driver.cc:271] failed call to cuInit: CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected

1. Construa um QCNN

1.1 Montar circuitos em um gráfico TensorFlow

O TensorFlow Quantum (TFQ) fornece classes de camada projetadas para construção de circuitos em grafos. Um exemplo é a camada tfq.layers.AddCircuit que herda de tf.keras.Layer . Essa camada pode ser anexada ou anexada ao lote de circuitos de entrada, conforme mostrado na figura a seguir.

O snippet a seguir usa essa camada:

qubit = cirq.GridQubit(0, 0)

# Define some circuits.
circuit1 = cirq.Circuit(cirq.X(qubit))
circuit2 = cirq.Circuit(cirq.H(qubit))

# Convert to a tensor.
input_circuit_tensor = tfq.convert_to_tensor([circuit1, circuit2])

# Define a circuit that we want to append
y_circuit = cirq.Circuit(cirq.Y(qubit))

# Instantiate our layer
y_appender = tfq.layers.AddCircuit()

# Run our circuit tensor through the layer and save the output.
output_circuit_tensor = y_appender(input_circuit_tensor, append=y_circuit)

Examine o tensor de entrada:

print(tfq.from_tensor(input_circuit_tensor))
[cirq.Circuit([
     cirq.Moment(
         cirq.X(cirq.GridQubit(0, 0)),
     ),
 ])
 cirq.Circuit([
     cirq.Moment(
         cirq.H(cirq.GridQubit(0, 0)),
     ),
 ])                                   ]

E examine o tensor de saída:

print(tfq.from_tensor(output_circuit_tensor))
[cirq.Circuit([
     cirq.Moment(
         cirq.X(cirq.GridQubit(0, 0)),
     ),
     cirq.Moment(
         cirq.Y(cirq.GridQubit(0, 0)),
     ),
 ])
 cirq.Circuit([
     cirq.Moment(
         cirq.H(cirq.GridQubit(0, 0)),
     ),
     cirq.Moment(
         cirq.Y(cirq.GridQubit(0, 0)),
     ),
 ])                                   ]

Embora seja possível executar os exemplos abaixo sem usar tfq.layers.AddCircuit , é uma boa oportunidade para entender como funcionalidades complexas podem ser incorporadas aos gráficos de computação do TensorFlow.

1.2 Visão geral do problema

Você preparará um estado de cluster e treinará um classificador quântico para detectar se está "excitado" ou não. O estado do cluster é altamente emaranhado, mas não necessariamente difícil para um computador clássico. Para maior clareza, este é um conjunto de dados mais simples do que o usado no artigo.

Para esta tarefa de classificação, você implementará uma arquitetura QCNN profunda do tipo MERA , pois:

  1. Como o QCNN, o estado do cluster em um anel é translacionalmente invariável.
  2. O estado do cluster é altamente emaranhado.

Essa arquitetura deve ser eficaz na redução do emaranhamento, obtendo a classificação lendo um único qubit.

Um estado de cluster "excitado" é definido como um estado de cluster que teve uma porta cirq.rx aplicada a qualquer um de seus qubits. Qconv e QPool são discutidos posteriormente neste tutorial.

1.3 Blocos de construção do TensorFlow

Uma maneira de resolver esse problema com o TensorFlow Quantum é implementar o seguinte:

  1. A entrada para o modelo é um tensor de circuito - um circuito vazio ou uma porta X em um qubit específico indicando uma excitação.
  2. Os demais componentes quânticos do modelo são construídos com camadas tfq.layers.AddCircuit .
  3. Para inferência é usada uma camada tfq.layers.PQC . Isso lê \(\langle \hat{Z} \rangle\) e o compara com um rótulo de 1 para um estado excitado ou -1 para um estado não excitado.

1.4 Dados

Antes de construir seu modelo, você pode gerar seus dados. Nesse caso, serão excitações para o estado do cluster (o artigo original usa um conjunto de dados mais complicado). As excitações são representadas com portas cirq.rx Uma rotação grande o suficiente é considerada uma excitação e é rotulada como 1 e uma rotação que não é grande o suficiente é rotulada -1 e considerada não uma excitação.

def generate_data(qubits):
    """Generate training and testing data."""
    n_rounds = 20  # Produces n_rounds * n_qubits datapoints.
    excitations = []
    labels = []
    for n in range(n_rounds):
        for bit in qubits:
            rng = np.random.uniform(-np.pi, np.pi)
            excitations.append(cirq.Circuit(cirq.rx(rng)(bit)))
            labels.append(1 if (-np.pi / 2) <= rng <= (np.pi / 2) else -1)

    split_ind = int(len(excitations) * 0.7)
    train_excitations = excitations[:split_ind]
    test_excitations = excitations[split_ind:]

    train_labels = labels[:split_ind]
    test_labels = labels[split_ind:]

    return tfq.convert_to_tensor(train_excitations), np.array(train_labels), \
        tfq.convert_to_tensor(test_excitations), np.array(test_labels)

Você pode ver que, assim como no aprendizado de máquina comum, você cria um conjunto de treinamento e teste para usar para comparar o modelo. Você pode ver rapidamente alguns pontos de dados com:

sample_points, sample_labels, _, __ = generate_data(cirq.GridQubit.rect(1, 4))
print('Input:', tfq.from_tensor(sample_points)[0], 'Output:', sample_labels[0])
print('Input:', tfq.from_tensor(sample_points)[1], 'Output:', sample_labels[1])
Input: (0, 0): ───X^0.449─── Output: 1
Input: (0, 1): ───X^-0.74─── Output: -1

1.5 Definir camadas

Agora defina as camadas mostradas na figura acima no TensorFlow.

1.5.1 Estado do cluster

A primeira etapa é definir o estado do cluster usando Cirq , uma estrutura fornecida pelo Google para programar circuitos quânticos. Como esta é uma parte estática do modelo, incorpore-a usando a funcionalidade tfq.layers.AddCircuit .

def cluster_state_circuit(bits):
    """Return a cluster state on the qubits in `bits`."""
    circuit = cirq.Circuit()
    circuit.append(cirq.H.on_each(bits))
    for this_bit, next_bit in zip(bits, bits[1:] + [bits[0]]):
        circuit.append(cirq.CZ(this_bit, next_bit))
    return circuit

Exiba um circuito de estado de cluster para um retângulo de cirq.GridQubit s:

SVGCircuit(cluster_state_circuit(cirq.GridQubit.rect(1, 4)))
findfont: Font family ['Arial'] not found. Falling back to DejaVu Sans.

svg

1.5.2 camadas QCNN

Defina as camadas que compõem o modelo usando o papel Cong e Lukin QCNN . Existem alguns pré-requisitos:

  • As matrizes unitárias parametrizadas de um e dois qubits do artigo de Tucci .
  • Uma operação geral de pool de dois qubits parametrizada.
def one_qubit_unitary(bit, symbols):
    """Make a Cirq circuit enacting a rotation of the bloch sphere about the X,
    Y and Z axis, that depends on the values in `symbols`.
    """
    return cirq.Circuit(
        cirq.X(bit)**symbols[0],
        cirq.Y(bit)**symbols[1],
        cirq.Z(bit)**symbols[2])


def two_qubit_unitary(bits, symbols):
    """Make a Cirq circuit that creates an arbitrary two qubit unitary."""
    circuit = cirq.Circuit()
    circuit += one_qubit_unitary(bits[0], symbols[0:3])
    circuit += one_qubit_unitary(bits[1], symbols[3:6])
    circuit += [cirq.ZZ(*bits)**symbols[6]]
    circuit += [cirq.YY(*bits)**symbols[7]]
    circuit += [cirq.XX(*bits)**symbols[8]]
    circuit += one_qubit_unitary(bits[0], symbols[9:12])
    circuit += one_qubit_unitary(bits[1], symbols[12:])
    return circuit


def two_qubit_pool(source_qubit, sink_qubit, symbols):
    """Make a Cirq circuit to do a parameterized 'pooling' operation, which
    attempts to reduce entanglement down from two qubits to just one."""
    pool_circuit = cirq.Circuit()
    sink_basis_selector = one_qubit_unitary(sink_qubit, symbols[0:3])
    source_basis_selector = one_qubit_unitary(source_qubit, symbols[3:6])
    pool_circuit.append(sink_basis_selector)
    pool_circuit.append(source_basis_selector)
    pool_circuit.append(cirq.CNOT(control=source_qubit, target=sink_qubit))
    pool_circuit.append(sink_basis_selector**-1)
    return pool_circuit

Para ver o que você criou, imprima o circuito unitário de um qubit:

SVGCircuit(one_qubit_unitary(cirq.GridQubit(0, 0), sympy.symbols('x0:3')))

svg

E o circuito unitário de dois qubits:

SVGCircuit(two_qubit_unitary(cirq.GridQubit.rect(1, 2), sympy.symbols('x0:15')))

svg

E o circuito de pool de dois qubits:

SVGCircuit(two_qubit_pool(*cirq.GridQubit.rect(1, 2), sympy.symbols('x0:6')))

svg

1.5.2.1 Convolução quântica

Como no artigo de Cong e Lukin , defina a convolução quântica 1D como a aplicação de uma unidade parametrizada de dois qubits a cada par de qubits adjacentes com um passo de um.

def quantum_conv_circuit(bits, symbols):
    """Quantum Convolution Layer following the above diagram.
    Return a Cirq circuit with the cascade of `two_qubit_unitary` applied
    to all pairs of qubits in `bits` as in the diagram above.
    """
    circuit = cirq.Circuit()
    for first, second in zip(bits[0::2], bits[1::2]):
        circuit += two_qubit_unitary([first, second], symbols)
    for first, second in zip(bits[1::2], bits[2::2] + [bits[0]]):
        circuit += two_qubit_unitary([first, second], symbols)
    return circuit

Exiba o circuito (muito horizontal):

SVGCircuit(
    quantum_conv_circuit(cirq.GridQubit.rect(1, 8), sympy.symbols('x0:15')))

svg

1.5.2.2 Agrupamento quântico

Uma camada de pool quântico agrupa de qubits l10n \(N\) para qubits \(\frac{N}{2}\) usando o pool de dois qubits definido acima.

def quantum_pool_circuit(source_bits, sink_bits, symbols):
    """A layer that specifies a quantum pooling operation.
    A Quantum pool tries to learn to pool the relevant information from two
    qubits onto 1.
    """
    circuit = cirq.Circuit()
    for source, sink in zip(source_bits, sink_bits):
        circuit += two_qubit_pool(source, sink, symbols)
    return circuit

Examine um circuito de componente de pooling:

test_bits = cirq.GridQubit.rect(1, 8)

SVGCircuit(
    quantum_pool_circuit(test_bits[:4], test_bits[4:], sympy.symbols('x0:6')))

svg

1.6 Definição do modelo

Agora use as camadas definidas para construir uma CNN puramente quântica. Comece com oito qubits, agrupe até um e, em seguida, meça \(\langle \hat{Z} \rangle\).

def create_model_circuit(qubits):
    """Create sequence of alternating convolution and pooling operators 
    which gradually shrink over time."""
    model_circuit = cirq.Circuit()
    symbols = sympy.symbols('qconv0:63')
    # Cirq uses sympy.Symbols to map learnable variables. TensorFlow Quantum
    # scans incoming circuits and replaces these with TensorFlow variables.
    model_circuit += quantum_conv_circuit(qubits, symbols[0:15])
    model_circuit += quantum_pool_circuit(qubits[:4], qubits[4:],
                                          symbols[15:21])
    model_circuit += quantum_conv_circuit(qubits[4:], symbols[21:36])
    model_circuit += quantum_pool_circuit(qubits[4:6], qubits[6:],
                                          symbols[36:42])
    model_circuit += quantum_conv_circuit(qubits[6:], symbols[42:57])
    model_circuit += quantum_pool_circuit([qubits[6]], [qubits[7]],
                                          symbols[57:63])
    return model_circuit


# Create our qubits and readout operators in Cirq.
cluster_state_bits = cirq.GridQubit.rect(1, 8)
readout_operators = cirq.Z(cluster_state_bits[-1])

# Build a sequential model enacting the logic in 1.3 of this notebook.
# Here you are making the static cluster state prep as a part of the AddCircuit and the
# "quantum datapoints" are coming in the form of excitation
excitation_input = tf.keras.Input(shape=(), dtype=tf.dtypes.string)
cluster_state = tfq.layers.AddCircuit()(
    excitation_input, prepend=cluster_state_circuit(cluster_state_bits))

quantum_model = tfq.layers.PQC(create_model_circuit(cluster_state_bits),
                               readout_operators)(cluster_state)

qcnn_model = tf.keras.Model(inputs=[excitation_input], outputs=[quantum_model])

# Show the keras plot of the model
tf.keras.utils.plot_model(qcnn_model,
                          show_shapes=True,
                          show_layer_names=False,
                          dpi=70)

png

1.7 Treine o modelo

Treine o modelo em todo o lote para simplificar este exemplo.

# Generate some training data.
train_excitations, train_labels, test_excitations, test_labels = generate_data(
    cluster_state_bits)


# Custom accuracy metric.
@tf.function
def custom_accuracy(y_true, y_pred):
    y_true = tf.squeeze(y_true)
    y_pred = tf.map_fn(lambda x: 1.0 if x >= 0 else -1.0, y_pred)
    return tf.keras.backend.mean(tf.keras.backend.equal(y_true, y_pred))


qcnn_model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.02),
                   loss=tf.losses.mse,
                   metrics=[custom_accuracy])

history = qcnn_model.fit(x=train_excitations,
                         y=train_labels,
                         batch_size=16,
                         epochs=25,
                         verbose=1,
                         validation_data=(test_excitations, test_labels))
Epoch 1/25
7/7 [==============================] - 2s 176ms/step - loss: 0.8961 - custom_accuracy: 0.7143 - val_loss: 0.8012 - val_custom_accuracy: 0.7500
Epoch 2/25
7/7 [==============================] - 1s 140ms/step - loss: 0.7736 - custom_accuracy: 0.7946 - val_loss: 0.7355 - val_custom_accuracy: 0.8542
Epoch 3/25
7/7 [==============================] - 1s 138ms/step - loss: 0.7319 - custom_accuracy: 0.8393 - val_loss: 0.7045 - val_custom_accuracy: 0.8125
Epoch 4/25
7/7 [==============================] - 1s 137ms/step - loss: 0.6976 - custom_accuracy: 0.8482 - val_loss: 0.6829 - val_custom_accuracy: 0.8333
Epoch 5/25
7/7 [==============================] - 1s 143ms/step - loss: 0.6696 - custom_accuracy: 0.8750 - val_loss: 0.6749 - val_custom_accuracy: 0.7917
Epoch 6/25
7/7 [==============================] - 1s 137ms/step - loss: 0.6631 - custom_accuracy: 0.8750 - val_loss: 0.6718 - val_custom_accuracy: 0.7917
Epoch 7/25
7/7 [==============================] - 1s 135ms/step - loss: 0.6536 - custom_accuracy: 0.8929 - val_loss: 0.6638 - val_custom_accuracy: 0.8750
Epoch 8/25
7/7 [==============================] - 1s 141ms/step - loss: 0.6376 - custom_accuracy: 0.8750 - val_loss: 0.6311 - val_custom_accuracy: 0.8542
Epoch 9/25
7/7 [==============================] - 1s 137ms/step - loss: 0.6208 - custom_accuracy: 0.8750 - val_loss: 0.5995 - val_custom_accuracy: 0.8542
Epoch 10/25
7/7 [==============================] - 1s 134ms/step - loss: 0.5887 - custom_accuracy: 0.8661 - val_loss: 0.5655 - val_custom_accuracy: 0.8333
Epoch 11/25
7/7 [==============================] - 1s 144ms/step - loss: 0.5796 - custom_accuracy: 0.8482 - val_loss: 0.5681 - val_custom_accuracy: 0.8333
Epoch 12/25
7/7 [==============================] - 1s 143ms/step - loss: 0.5630 - custom_accuracy: 0.7946 - val_loss: 0.5179 - val_custom_accuracy: 0.8333
Epoch 13/25
7/7 [==============================] - 1s 137ms/step - loss: 0.5405 - custom_accuracy: 0.8304 - val_loss: 0.5003 - val_custom_accuracy: 0.8333
Epoch 14/25
7/7 [==============================] - 1s 138ms/step - loss: 0.5259 - custom_accuracy: 0.8036 - val_loss: 0.4787 - val_custom_accuracy: 0.8333
Epoch 15/25
7/7 [==============================] - 1s 137ms/step - loss: 0.5077 - custom_accuracy: 0.8482 - val_loss: 0.4741 - val_custom_accuracy: 0.8125
Epoch 16/25
7/7 [==============================] - 1s 136ms/step - loss: 0.5082 - custom_accuracy: 0.8214 - val_loss: 0.4739 - val_custom_accuracy: 0.8125
Epoch 17/25
7/7 [==============================] - 1s 137ms/step - loss: 0.5138 - custom_accuracy: 0.8214 - val_loss: 0.4859 - val_custom_accuracy: 0.8750
Epoch 18/25
7/7 [==============================] - 1s 133ms/step - loss: 0.5073 - custom_accuracy: 0.8304 - val_loss: 0.4879 - val_custom_accuracy: 0.8333
Epoch 19/25
7/7 [==============================] - 1s 138ms/step - loss: 0.5084 - custom_accuracy: 0.8304 - val_loss: 0.4745 - val_custom_accuracy: 0.8542
Epoch 20/25
7/7 [==============================] - 1s 139ms/step - loss: 0.5057 - custom_accuracy: 0.8571 - val_loss: 0.4702 - val_custom_accuracy: 0.8333
Epoch 21/25
7/7 [==============================] - 1s 135ms/step - loss: 0.4939 - custom_accuracy: 0.8304 - val_loss: 0.4734 - val_custom_accuracy: 0.8750
Epoch 22/25
7/7 [==============================] - 1s 138ms/step - loss: 0.4942 - custom_accuracy: 0.8750 - val_loss: 0.4725 - val_custom_accuracy: 0.8750
Epoch 23/25
7/7 [==============================] - 1s 140ms/step - loss: 0.4982 - custom_accuracy: 0.9107 - val_loss: 0.4695 - val_custom_accuracy: 0.8958
Epoch 24/25
7/7 [==============================] - 1s 135ms/step - loss: 0.4936 - custom_accuracy: 0.8661 - val_loss: 0.4731 - val_custom_accuracy: 0.8750
Epoch 25/25
7/7 [==============================] - 1s 136ms/step - loss: 0.4866 - custom_accuracy: 0.8571 - val_loss: 0.4631 - val_custom_accuracy: 0.8958
plt.plot(history.history['loss'][1:], label='Training')
plt.plot(history.history['val_loss'][1:], label='Validation')
plt.title('Training a Quantum CNN to Detect Excited Cluster States')
plt.xlabel('Epochs')
plt.ylabel('Loss')
plt.legend()
plt.show()

png

2. Modelos híbridos

Você não precisa ir de oito qubits para um qubit usando convolução quântica – você poderia ter feito uma ou duas rodadas de convolução quântica e alimentado os resultados em uma rede neural clássica. Esta seção explora modelos híbridos quânticos-clássicos.

2.1 Modelo híbrido com um único filtro quântico

Aplique uma camada de convolução quântica, lendo \(\langle \hat{Z}_n \rangle\) em todos os bits, seguido por uma rede neural densamente conectada.

2.1.1 Definição do modelo

# 1-local operators to read out
readouts = [cirq.Z(bit) for bit in cluster_state_bits[4:]]


def multi_readout_model_circuit(qubits):
    """Make a model circuit with less quantum pool and conv operations."""
    model_circuit = cirq.Circuit()
    symbols = sympy.symbols('qconv0:21')
    model_circuit += quantum_conv_circuit(qubits, symbols[0:15])
    model_circuit += quantum_pool_circuit(qubits[:4], qubits[4:],
                                          symbols[15:21])
    return model_circuit


# Build a model enacting the logic in 2.1 of this notebook.
excitation_input_dual = tf.keras.Input(shape=(), dtype=tf.dtypes.string)

cluster_state_dual = tfq.layers.AddCircuit()(
    excitation_input_dual, prepend=cluster_state_circuit(cluster_state_bits))

quantum_model_dual = tfq.layers.PQC(
    multi_readout_model_circuit(cluster_state_bits),
    readouts)(cluster_state_dual)

d1_dual = tf.keras.layers.Dense(8)(quantum_model_dual)

d2_dual = tf.keras.layers.Dense(1)(d1_dual)

hybrid_model = tf.keras.Model(inputs=[excitation_input_dual], outputs=[d2_dual])

# Display the model architecture
tf.keras.utils.plot_model(hybrid_model,
                          show_shapes=True,
                          show_layer_names=False,
                          dpi=70)

png

2.1.2 Treinar o modelo

hybrid_model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.02),
                     loss=tf.losses.mse,
                     metrics=[custom_accuracy])

hybrid_history = hybrid_model.fit(x=train_excitations,
                                  y=train_labels,
                                  batch_size=16,
                                  epochs=25,
                                  verbose=1,
                                  validation_data=(test_excitations,
                                                   test_labels))
Epoch 1/25
7/7 [==============================] - 1s 113ms/step - loss: 0.9848 - custom_accuracy: 0.5179 - val_loss: 0.9635 - val_custom_accuracy: 0.5417
Epoch 2/25
7/7 [==============================] - 1s 86ms/step - loss: 0.8095 - custom_accuracy: 0.6339 - val_loss: 0.6800 - val_custom_accuracy: 0.7083
Epoch 3/25
7/7 [==============================] - 1s 85ms/step - loss: 0.4045 - custom_accuracy: 0.9375 - val_loss: 0.3342 - val_custom_accuracy: 0.8750
Epoch 4/25
7/7 [==============================] - 1s 86ms/step - loss: 0.2308 - custom_accuracy: 0.9643 - val_loss: 0.2027 - val_custom_accuracy: 0.9792
Epoch 5/25
7/7 [==============================] - 1s 84ms/step - loss: 0.2232 - custom_accuracy: 0.9554 - val_loss: 0.1761 - val_custom_accuracy: 1.0000
Epoch 6/25
7/7 [==============================] - 1s 84ms/step - loss: 0.1760 - custom_accuracy: 0.9821 - val_loss: 0.2541 - val_custom_accuracy: 0.9167
Epoch 7/25
7/7 [==============================] - 1s 85ms/step - loss: 0.1919 - custom_accuracy: 0.9643 - val_loss: 0.1967 - val_custom_accuracy: 0.9792
Epoch 8/25
7/7 [==============================] - 1s 83ms/step - loss: 0.1892 - custom_accuracy: 0.9554 - val_loss: 0.1870 - val_custom_accuracy: 0.9792
Epoch 9/25
7/7 [==============================] - 1s 84ms/step - loss: 0.1777 - custom_accuracy: 0.9911 - val_loss: 0.2208 - val_custom_accuracy: 0.9583
Epoch 10/25
7/7 [==============================] - 1s 83ms/step - loss: 0.1728 - custom_accuracy: 0.9732 - val_loss: 0.2147 - val_custom_accuracy: 0.9583
Epoch 11/25
7/7 [==============================] - 1s 85ms/step - loss: 0.1704 - custom_accuracy: 0.9732 - val_loss: 0.1810 - val_custom_accuracy: 0.9792
Epoch 12/25
7/7 [==============================] - 1s 85ms/step - loss: 0.1739 - custom_accuracy: 0.9732 - val_loss: 0.2038 - val_custom_accuracy: 0.9792
Epoch 13/25
7/7 [==============================] - 1s 81ms/step - loss: 0.1705 - custom_accuracy: 0.9732 - val_loss: 0.1855 - val_custom_accuracy: 0.9792
Epoch 14/25
7/7 [==============================] - 1s 84ms/step - loss: 0.1788 - custom_accuracy: 0.9643 - val_loss: 0.2152 - val_custom_accuracy: 0.9583
Epoch 15/25
7/7 [==============================] - 1s 84ms/step - loss: 0.1760 - custom_accuracy: 0.9732 - val_loss: 0.1994 - val_custom_accuracy: 1.0000
Epoch 16/25
7/7 [==============================] - 1s 83ms/step - loss: 0.1737 - custom_accuracy: 0.9732 - val_loss: 0.2035 - val_custom_accuracy: 0.9792
Epoch 17/25
7/7 [==============================] - 1s 82ms/step - loss: 0.1749 - custom_accuracy: 0.9911 - val_loss: 0.1983 - val_custom_accuracy: 0.9583
Epoch 18/25
7/7 [==============================] - 1s 83ms/step - loss: 0.1875 - custom_accuracy: 0.9732 - val_loss: 0.1916 - val_custom_accuracy: 0.9583
Epoch 19/25
7/7 [==============================] - 1s 82ms/step - loss: 0.1605 - custom_accuracy: 0.9732 - val_loss: 0.1782 - val_custom_accuracy: 0.9792
Epoch 20/25
7/7 [==============================] - 1s 84ms/step - loss: 0.1668 - custom_accuracy: 0.9911 - val_loss: 0.2276 - val_custom_accuracy: 0.9583
Epoch 21/25
7/7 [==============================] - 1s 84ms/step - loss: 0.1700 - custom_accuracy: 0.9911 - val_loss: 0.2080 - val_custom_accuracy: 0.9583
Epoch 22/25
7/7 [==============================] - 1s 83ms/step - loss: 0.1621 - custom_accuracy: 0.9732 - val_loss: 0.1851 - val_custom_accuracy: 0.9375
Epoch 23/25
7/7 [==============================] - 1s 84ms/step - loss: 0.1695 - custom_accuracy: 0.9911 - val_loss: 0.1882 - val_custom_accuracy: 0.9792
Epoch 24/25
7/7 [==============================] - 1s 82ms/step - loss: 0.1583 - custom_accuracy: 0.9911 - val_loss: 0.2017 - val_custom_accuracy: 0.9583
Epoch 25/25
7/7 [==============================] - 1s 83ms/step - loss: 0.1557 - custom_accuracy: 0.9911 - val_loss: 0.1907 - val_custom_accuracy: 0.9792
plt.plot(history.history['val_custom_accuracy'], label='QCNN')
plt.plot(hybrid_history.history['val_custom_accuracy'], label='Hybrid CNN')
plt.title('Quantum vs Hybrid CNN performance')
plt.xlabel('Epochs')
plt.legend()
plt.ylabel('Validation Accuracy')
plt.show()

png

Como você pode ver, com uma assistência clássica muito modesta, o modelo híbrido geralmente converge mais rápido do que a versão puramente quântica.

2.2 Convolução híbrida com vários filtros quânticos

Agora vamos tentar uma arquitetura que usa várias convoluções quânticas e uma rede neural clássica para combiná-las.

2.2.1 Definição do modelo

excitation_input_multi = tf.keras.Input(shape=(), dtype=tf.dtypes.string)

cluster_state_multi = tfq.layers.AddCircuit()(
    excitation_input_multi, prepend=cluster_state_circuit(cluster_state_bits))

# apply 3 different filters and measure expectation values

quantum_model_multi1 = tfq.layers.PQC(
    multi_readout_model_circuit(cluster_state_bits),
    readouts)(cluster_state_multi)

quantum_model_multi2 = tfq.layers.PQC(
    multi_readout_model_circuit(cluster_state_bits),
    readouts)(cluster_state_multi)

quantum_model_multi3 = tfq.layers.PQC(
    multi_readout_model_circuit(cluster_state_bits),
    readouts)(cluster_state_multi)

# concatenate outputs and feed into a small classical NN
concat_out = tf.keras.layers.concatenate(
    [quantum_model_multi1, quantum_model_multi2, quantum_model_multi3])

dense_1 = tf.keras.layers.Dense(8)(concat_out)

dense_2 = tf.keras.layers.Dense(1)(dense_1)

multi_qconv_model = tf.keras.Model(inputs=[excitation_input_multi],
                                   outputs=[dense_2])

# Display the model architecture
tf.keras.utils.plot_model(multi_qconv_model,
                          show_shapes=True,
                          show_layer_names=True,
                          dpi=70)

png

2.2.2 Treinar o modelo

multi_qconv_model.compile(
    optimizer=tf.keras.optimizers.Adam(learning_rate=0.02),
    loss=tf.losses.mse,
    metrics=[custom_accuracy])

multi_qconv_history = multi_qconv_model.fit(x=train_excitations,
                                            y=train_labels,
                                            batch_size=16,
                                            epochs=25,
                                            verbose=1,
                                            validation_data=(test_excitations,
                                                             test_labels))
Epoch 1/25
7/7 [==============================] - 2s 143ms/step - loss: 0.9425 - custom_accuracy: 0.6429 - val_loss: 0.8120 - val_custom_accuracy: 0.7083
Epoch 2/25
7/7 [==============================] - 1s 109ms/step - loss: 0.5778 - custom_accuracy: 0.7946 - val_loss: 0.5920 - val_custom_accuracy: 0.7500
Epoch 3/25
7/7 [==============================] - 1s 103ms/step - loss: 0.4954 - custom_accuracy: 0.9018 - val_loss: 0.4568 - val_custom_accuracy: 0.7708
Epoch 4/25
7/7 [==============================] - 1s 95ms/step - loss: 0.2855 - custom_accuracy: 0.9196 - val_loss: 0.2792 - val_custom_accuracy: 0.9375
Epoch 5/25
7/7 [==============================] - 1s 93ms/step - loss: 0.1902 - custom_accuracy: 0.9821 - val_loss: 0.2212 - val_custom_accuracy: 0.9375
Epoch 6/25
7/7 [==============================] - 1s 94ms/step - loss: 0.1685 - custom_accuracy: 0.9821 - val_loss: 0.2341 - val_custom_accuracy: 0.9583
Epoch 7/25
7/7 [==============================] - 1s 104ms/step - loss: 0.1671 - custom_accuracy: 0.9911 - val_loss: 0.2062 - val_custom_accuracy: 0.9792
Epoch 8/25
7/7 [==============================] - 1s 97ms/step - loss: 0.1511 - custom_accuracy: 0.9821 - val_loss: 0.2096 - val_custom_accuracy: 0.9792
Epoch 9/25
7/7 [==============================] - 1s 96ms/step - loss: 0.1432 - custom_accuracy: 0.9911 - val_loss: 0.2330 - val_custom_accuracy: 0.9375
Epoch 10/25
7/7 [==============================] - 1s 92ms/step - loss: 0.1668 - custom_accuracy: 0.9821 - val_loss: 0.2344 - val_custom_accuracy: 0.9583
Epoch 11/25
7/7 [==============================] - 1s 106ms/step - loss: 0.1893 - custom_accuracy: 0.9732 - val_loss: 0.2148 - val_custom_accuracy: 0.9583
Epoch 12/25
7/7 [==============================] - 1s 104ms/step - loss: 0.1857 - custom_accuracy: 0.9732 - val_loss: 0.2739 - val_custom_accuracy: 0.9583
Epoch 13/25
7/7 [==============================] - 1s 106ms/step - loss: 0.1748 - custom_accuracy: 0.9732 - val_loss: 0.2366 - val_custom_accuracy: 0.9583
Epoch 14/25
7/7 [==============================] - 1s 103ms/step - loss: 0.1515 - custom_accuracy: 0.9821 - val_loss: 0.2012 - val_custom_accuracy: 0.9583
Epoch 15/25
7/7 [==============================] - 1s 100ms/step - loss: 0.1552 - custom_accuracy: 0.9911 - val_loss: 0.2404 - val_custom_accuracy: 0.9375
Epoch 16/25
7/7 [==============================] - 1s 97ms/step - loss: 0.1572 - custom_accuracy: 0.9911 - val_loss: 0.2779 - val_custom_accuracy: 0.9375
Epoch 17/25
7/7 [==============================] - 1s 100ms/step - loss: 0.1546 - custom_accuracy: 0.9821 - val_loss: 0.2104 - val_custom_accuracy: 0.9583
Epoch 18/25
7/7 [==============================] - 1s 102ms/step - loss: 0.1418 - custom_accuracy: 0.9911 - val_loss: 0.2647 - val_custom_accuracy: 0.9583
Epoch 19/25
7/7 [==============================] - 1s 98ms/step - loss: 0.1590 - custom_accuracy: 0.9732 - val_loss: 0.2154 - val_custom_accuracy: 0.9583
Epoch 20/25
7/7 [==============================] - 1s 104ms/step - loss: 0.1363 - custom_accuracy: 1.0000 - val_loss: 0.2470 - val_custom_accuracy: 0.9375
Epoch 21/25
7/7 [==============================] - 1s 100ms/step - loss: 0.1442 - custom_accuracy: 0.9821 - val_loss: 0.2383 - val_custom_accuracy: 0.9375
Epoch 22/25
7/7 [==============================] - 1s 99ms/step - loss: 0.1415 - custom_accuracy: 0.9911 - val_loss: 0.2324 - val_custom_accuracy: 0.9583
Epoch 23/25
7/7 [==============================] - 1s 97ms/step - loss: 0.1424 - custom_accuracy: 0.9821 - val_loss: 0.2188 - val_custom_accuracy: 0.9583
Epoch 24/25
7/7 [==============================] - 1s 100ms/step - loss: 0.1417 - custom_accuracy: 0.9821 - val_loss: 0.2340 - val_custom_accuracy: 0.9375
Epoch 25/25
7/7 [==============================] - 1s 103ms/step - loss: 0.1471 - custom_accuracy: 0.9732 - val_loss: 0.2252 - val_custom_accuracy: 0.9583
plt.plot(history.history['val_custom_accuracy'][:25], label='QCNN')
plt.plot(hybrid_history.history['val_custom_accuracy'][:25], label='Hybrid CNN')
plt.plot(multi_qconv_history.history['val_custom_accuracy'][:25],
         label='Hybrid CNN \n Multiple Quantum Filters')
plt.title('Quantum vs Hybrid CNN performance')
plt.xlabel('Epochs')
plt.legend()
plt.ylabel('Validation Accuracy')
plt.show()

png