Millions of activations from an image classification #AI #neuralNetwork has created an “Activation Atlas” that shows how it analyzes an image to make sense of it – akin to looking “into” someone’s …

By using feature inversion to visualize millions of activations from an image classification network, we create an explorable activation atlas of features the network has learned which can reveal how the network typically represents some concepts.

Sourced through Scoop.it from: distill.pub

WHY IT MATTERS: this very technical paper provides some easy to understand visualizations of the way a neural network analyzes images…

Tags:
,
Farid Mheir
farid@mheir.com