Developing and training the AlexNet model using Tensorflow on CIFAR-10 dataset

1. import libraries

import tensorflow as tf
from tensorflow.keras import datasets, layers, models, optimizers, regularizers

 

2. Load and preprocess CIFAR-10 dataset

(train_images, train_labels), (test_images, test_labels) = datasets.cifar10.load_data()
train_images, test_images = train_images / 255.0, test_images / 255.0

num_train = int(len(train_images) * 0.8)
train_images, validation_images = train_images[:num_train], train_images[num_train:]
train_labels, validation_labels = train_labels[:num_train], train_labels[num_train:]

 

3. Define AlexNet-like model

def create_alexnet():
    model = models.Sequential([
        layers.experimental.preprocessing.Resizing(224, 224, input_shape=(32, 32, 3)),

        layers.Conv2D(96, (11, 11), strides=(4, 4), activation='relu'),
        layers.BatchNormalization(),
        layers.MaxPooling2D((3, 3), strides=(2, 2)),

        layers.Conv2D(256, (5, 5), padding='same', activation='relu'),
        layers.BatchNormalization(),
        layers.MaxPooling2D((3, 3), strides=(2, 2)),

        layers.Conv2D(384, (3, 3), padding='same', activation='relu'),
        layers.BatchNormalization(),

        layers.Conv2D(384, (3, 3), padding='same', activation='relu'),
        layers.BatchNormalization(),

        layers.Conv2D(256, (3, 3), padding='same', activation='relu'),
        layers.BatchNormalization(),
        layers.MaxPooling2D((3, 3), strides=(2, 2)),

        layers.Flatten(),
        layers.Dense(4096, activation='relu', kernel_regularizer=regularizers.l2(0.0005)),
        layers.Dropout(0.5),

        layers.Dense(4096, activation='relu', kernel_regularizer=regularizers.l2(0.0005)),
        layers.Dropout(0.5),

        layers.Dense(10, activation='softmax')
    ])
    return model

 

4. Create the model

model = create_alexnet()

 

5. Compile the model

model.compile(optimizer=optimizers.Adam(learning_rate=0.001),
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])

 

6. Train the model

history = model.fit(train_images, train_labels, epochs=50, batch_size=128, validation_data=(validation_images, validation_labels))

 

7. Result

Epoch 1/50
313/313 [==============================] - 58s 125ms/step - loss: 6.3220 - accuracy: 0.3163 - val_loss: 4.0909 - val_accuracy: 0.4078
Epoch 2/50
313/313 [==============================] - 40s 127ms/step - loss: 3.3782 - accuracy: 0.4340 - val_loss: 3.1517 - val_accuracy: 0.3418
Epoch 3/50
313/313 [==============================] - 38s 122ms/step - loss: 2.5291 - accuracy: 0.5001 - val_loss: 2.5480 - val_accuracy: 0.4140
Epoch 4/50
313/313 [==============================] - 39s 125ms/step - loss: 2.1386 - accuracy: 0.5583 - val_loss: 3.2482 - val_accuracy: 0.3013
Epoch 5/50
313/313 [==============================] - 42s 133ms/step - loss: 1.9543 - accuracy: 0.6087 - val_loss: 2.6728 - val_accuracy: 0.4096
Epoch 6/50
313/313 [==============================] - 42s 135ms/step - loss: 1.7871 - accuracy: 0.6665 - val_loss: 1.8894 - val_accuracy: 0.6108

...

Epoch 47/50
313/313 [==============================] - 43s 137ms/step - loss: 0.2638 - accuracy: 0.9839 - val_loss: 1.2113 - val_accuracy: 0.8090
Epoch 48/50
313/313 [==============================] - 41s 130ms/step - loss: 0.2613 - accuracy: 0.9833 - val_loss: 1.3618 - val_accuracy: 0.7728
Epoch 49/50
313/313 [==============================] - 43s 136ms/step - loss: 0.2441 - accuracy: 0.9863 - val_loss: 1.1390 - val_accuracy: 0.8094
Epoch 50/50
313/313 [==============================] - 42s 136ms/step - loss: 0.2470 - accuracy: 0.9844 - val_loss: 1.2852 - val_accuracy: 0.7858