liveBook · Manning

Image recognition

As explained in Grokking Machine Learning, Second Edition

The computational task of identifying objects, patterns, or other meaningful content in images.

Image recognition turns image pixels into a finite-state prediction, such as identifying a handwritten digit or deciding which object appears in a picture. section 2.9

MNIST provides 28-by-28 grayscale images of digits 0–9. A neural network reshapes each image into 784 numbers and converts each digit label into a ten-entry indicator vector. section 10.19

A small image-recognition workflow:

A model’s mistake

The trained network correctly classifies one image as 4 but incorrectly classifies an image of 3 as 8, showing why testing matters.

How Grokking Machine Learning, Second Edition covers it

How deep it goes
Descends from classification intuition to reshaping, label encoding, neural-network training, and prediction inspection.
Examples
Uses MNIST handwritten digits and contrasts a correctly classified 4 with a 3 mistaken for an 8.
What you can do after
Builds and tests a Keras neural network that predicts handwritten digits from MNIST images.
How it connects
Connects classification with neural networks, MNIST, preprocessing, training, testing, and overfitting.

Also explained in

Compare how 3 books explain image recognition →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Grokking Machine Learning, Second Edition. The section links point to the parts of the book it draws on.