On MNIST Dataset

Faisal Qureshi
faisal.qureshi@ontariotechu.ca
http://www.vclab.ca

What is MNIST?

• MNIST (Modified National Institute of Standards and Technology) dataset consists of 70,000 grayscale images of handwritten digits (0-9).
• Each image is 28 $\times$ 28 pixels and labeled with a corresponding digit.
• Used extensively for benchmarking machine learning and deep learning models.

History of MNIST

• Originates from the NIST dataset used for US postal service digit recognition.
• Yann LeCun, Corinna Cortes, and Christopher Burges modified the dataset to create MNIST in 1998.
• Became a standard dataset for evaluating image classification algorithms.

MNIST Dataset Composition

• Training Set: 60,000 images
• Test Set: 10,000 images
• Classes: 10 (digits 0-9)
• Data Format:
  • Images: 28 $\times$ 28 pixel grayscale.
  • Labels: Single digit (0-9).

Benefits

• Small and Simple: Easy to train models quickly.
• Well-Structured: Standardized format enables consistent benchmarking.
• Broadly Used: Suitable for deep learning, feature extraction, and traditional machine learning models.

Using MNIST in Python

from torchvision import datasets, transforms
from torch.utils.data import DataLoader

# Load MNIST dataset
dataset = datasets.MNIST(root='../datasets/common', train=True, transform=transforms.ToTensor(), download=True)
dataloader = DataLoader(dataset, batch_size=32, shuffle=True)

Sample MNIST Images

import matplotlib.pyplot as plt
import numpy as np

# Load dataset
mnist = datasets.MNIST(root='../datasets/common', train=True, transform=transforms.ToTensor(), download=True)
images, labels = zip(*[mnist[i] for i in range(10)])

# Plot images
plt.suptitle('MNIST Samples')
fig, axes = plt.subplots(1, 10, figsize=(10, 2))
for i, ax in enumerate(axes):
    ax.imshow(images[i].numpy().squeeze(), cmap='gray')
    ax.set_title(labels[i])
    ax.axis('off')
plt.show()

<Figure size 640x480 with 0 Axes>

Applications of MNIST

• Benchmarking Image Classification Algorithms
• Feature Learning and Representation
• Handwriting Recognition Systems
• Experimenting with Neural Networks (CNNs, Autoencoders, GANs, VAEs)

MNIST Extensions

• Fashion-MNIST: Clothing images instead of digits.
• EMNIST: Extended MNIST with letters.
• KMNIST: Kuzushiji characters (Japanese handwriting dataset).

Conclusions

• MNIST remains a fundamental dataset for learning and benchmarking image classification models.
• Despite being solved by modern deep learning, it serves as an excellent starting point for understanding neural networks.