Why do I get AttributeError: module 'tensorflow.keras.backend' has no attribute 'placeholder' when using ART's KerasClassifier with TensorFlow 2.x?

I’m trying to use the Adversarial Robustness Toolbox (ART) with a simple Keras model in TensorFlow 2.x, but I’m encountering the following error: AttributeError: module 'tensorflow.keras.backend' has no attribute 'placeholder

My code looks like this:

import tensorflow as tf
print(tf.__version__)

import numpy as np
from art.estimators.classification import KerasClassifier
from art.attacks.evasion import FastGradientMethod
from art.utils import load_mnist
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Flatten
from tensorflow.keras.optimizers import Adam

# Load the MNIST-Dataset
(x_train, y_train), (x_test, y_test), min_, max_ = load_mnist()
x_train = x_train.reshape((x_train.shape[0], 28, 28, 1)).astype('float32') / 255.0
x_test = x_test.reshape((x_test.shape[0], 28, 28, 1)).astype('float32') / 255.0

# Create keras model
model = Sequential([
    Flatten(input_shape=(28, 28, 1)),
    Dense(128, activation='relu'),
    Dense(10, activation='softmax')
])
model.compile(optimizer=Adam(), loss='categorical_crossentropy', metrics=['accuracy'])

# Train the model
model.fit(x_train, y_train, epochs=3, batch_size=128, verbose=1)

tf.compat.v1.disable_eager_execution()
# Compile the keras modell into an ART classifiert
classifier = KerasClassifier(model=model, clip_values=(0, 1), use_logits=False)

# configure FGSM-attack
attack = FastGradientMethod(estimator=classifier, eps=0.2)

# create adversarial samples
x_test_adv = attack.generate(x=x_test)

# Evaluation of the model with adversarial samples
predictions = np.argmax(classifier.predict(x_test_adv), axis=1)
accuracy = np.sum(predictions == np.argmax(y_test, axis=1)) / len(y_test)
print(f"Genauigkeit nach Angriff: {accuracy * 100:.2f}%")

# Visualisation
import matplotlib.pyplot as plt

plt.figure(figsize=(10, 5))
for i in range(5):
    plt.subplot(2, 5, i + 1)
    plt.imshow(x_test[i].reshape(28, 28), cmap="gray")
    plt.title("Original")
    plt.axis("off")
    
    plt.subplot(2, 5, i + 6)
    plt.imshow(x_test_adv[i].reshape(28, 28), cmap="gray")
    plt.title("Adversarial")
    plt.axis("off")
plt.tight_layout()
plt.show()

This error is raised within the ART library when initializing the KerasClassifier. I have TensorFlow 2.x installed, and I’m using the latest version of the Adversarial Robustness Toolbox (ART).

Things I’ve tried:

• Disabling eager execution using tf.compat.v1.disable_eager_execution(), but it doesn't solve the issue.
• Checking if there are any updates for ART or TensorFlow, but everything is up-to-date.

Can anyone suggest why this error occurs and how to resolve it in TensorFlow 2.x?