> ML_DATASET // MNIST-DIGITS_v1.0
MNIST Handwritten Digits (LeCun, Cortes & Burges 1998)
Yann LeCun / Corinna Cortes / Christopher Burges (NYU / Bell Labs) · Optical Character Recognition & Education · 70,000 28x28 grayscale images of handwritten digits 0-9
Optical Character Recognition & EducationCC-BY-SA-3.070,000 28x28 grayscale images of handwritten digits 0-9open
Dataset Profile & Characteristics
Label Type:Single digit label (0 to 9)
Languages:en
License Tier:copyleft
Modalities:image
Intended Use
- Educational "Hello World" of deep learning
- Verifying gradient backpropagation implementations
Prohibited / Discouraged Use
- Evaluating novel state-of-the-art architectures in 2026
Bias, Leakage & Privacy Risk Analysis
Privacy / Sensitive Data Risks:
High-school students and Census Bureau employee handwritten digits from 1990s.
Known Bias:
Clean centered binary digits; trivial to achieve >99.7% accuracy.
Known Benchmark Leakage:
Standard 60,000 train / 10,000 test split.