Member-only story
Tensorflow vs PyTorch by Example
How to apply each deep learning framework to the MNIST problem
In the late 1990s, machine learning researchers were experimenting with ways to create artificial neural networks in layered architectures that could perform simple computer vision tasks. Machine learning is the science of writing algorithms that can learn from data.
One classic benchmark of computer vision algorithms is their performance against the MNIST database.
This database, whose name stands for Modified National Institute of Standards and Technology, consists of 70,000 small pictures of hand-written digits, each represented as a 28x28 pixel grid of grayscale color values.
Each image is tagged with the number it represents. The challenge for the algorithm is to do what comes easily to humans: look at the picture and tell us what number it is.
The 1998 paper by Yann Lecun and collaborators introduced the world to the MNIST database and showed that convolutional neural networks (CNNs) outperformed other techniques. The LeNet-5 architecture achieved an error rate of just 0.8%.
In the years since, convolutional networks are still one of the preferred techniques used in state-of-the-art models, with published error rates getting down to around 0.2%.
