saadselim.com

Many decades ago, artificial neural networks were developed to mimic the learning capabilities of humans and animals. Below is an excerpt from The Machine that Changed the World, a 1992 documentary about Artificial Intelligence.

from IPython.display import YouTubeVideo
YouTubeVideo('cNxadbrN_aI')

Since then, computers and machine learning libraries have evolved dramatically. This notebook will demonstrate how neural networks have improved and the biological inspiration behind them.

We'll use TensorFlow, a popular open-source library that automatically detects GPUs for computation.

import tensorflow as tf
tf.config.list_physical_devices('GPU')

Data

We'll tackle image classification using the Fashion MNIST dataset:

fashion_mnist = tf.keras.datasets.fashion_mnist
(train_images, train_labels), (valid_images, valid_labels) = fashion_mnist.load_data()

Visualizing an image:

import matplotlib.pyplot as plt
 
data_idx = 42
 
plt.figure()
plt.imshow(train_images[data_idx], cmap='gray')
plt.colorbar()
plt.grid(False)
plt.show()

Categories:

Label	Description
0	T-shirt/top
1	Trouser
2	Pullover
3	Dress
4	Coat
5	Sandal
6	Shirt
7	Sneaker
8	Bag
9	Ankle boot

Check your guess:

train_labels[data_idx]

Building a Neuron

Neurons are basic neural network components inspired by biological neurons.

Architecture

Biological neurons transmit electrical impulses, while artificial neurons use simplified mathematics. The basic equation used is:

y = w0x0 + w1x1 + ... + b

Each image (28x28 pixels) results in 784 weights.

Model creation using Keras:

number_of_classes = train_labels.max() + 1
 
model = tf.keras.Sequential([
    tf.keras.layers.Flatten(input_shape=(28, 28)),
    tf.keras.layers.Dense(number_of_classes)
])

Verify the model

model.summary()

Parameter calculation:

image_height, image_width = 28, 28
number_of_weights = image_height * image_width * number_of_classes

Initiating Training

Loss function and metrics setup:

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

Evaluating the Model

Training and validation:

history = model.fit(
    train_images,
    train_labels,
    epochs=5,
    verbose=True,
    validation_data=(valid_images, valid_labels)
)

Prediction

Make predictions on new data:

predictions = model.predict(train_images[0:10])

Visualize predictions:

data_idx = 8675
 
plt.figure()
plt.imshow(train_images[data_idx], cmap='gray')
plt.colorbar()
plt.grid(False)
plt.show()
 
x_values = range(number_of_classes)
 
plt.figure()
plt.bar(x_values, model.predict(train_images[data_idx:data_idx+1]).flatten())
plt.xticks(range(10))
plt.show()
 
print("Correct answer:", train_labels[data_idx])

Congratulations! You've built a neural network capable of basic image classification.