liveBook · Manning

OutlierAE

As explained in Outlier Detection in Python

OutlierAE is an alibi-detect detector class that identifies outliers using an autoencoder's reconstruction error.

OutlierAE is an alibi-detect detector that identifies unusual inputs through an autoencoder’s reconstruction error. It reconstructs typical data well; inputs reconstructed poorly receive higher anomaly scores. section 16.4

A practical image-detection workflow:

  • Fit the detector, then call predict() with instance and feature scores enabled. section 16.4
  • Interpret unusually large mean squared reconstruction errors as possible outliers rather than automatically meaningful anomalies. section 16.4

Noisy ship image

A typical ship image is reconstructed closely and is not flagged. After random pixel noise is added, reconstruction error rises and the image is flagged as an outlier.

How Outlier Detection in Python covers it

How deep it goes
Implements encoder-decoder construction, fitting, prediction, scoring, and deliberate image perturbation, but not detector theory.
Examples
Uses CIFAR10 images, a typical ship image, and the same image corrupted with random pixel noise.
What you can do after
Apply OutlierAE with Keras and alibi-detect to score image instances and inspect their reconstructions.
How it connects
Connects autoencoders, convolutional layers, reconstruction error, mean squared error, and image outlier detection.
Cover of Outlier Detection in Python
liveBookmanning.com
Read Outlier Detection in Python on liveBook Get the book at manning.com

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Outlier Detection in Python. The section links point to the parts of the book it draws on.