Visually Understanding Deep Neural Networks with the help of Dimensionality Reduction (EN)

Visually Understanding Deep Neural Networks with the help of Dimensionality Reduction (EN)

14:15 - 14:45 | tech.stage | Neural Networks

Given the (toy) example of visually comparing image trademarks we study the following approach:

Understanding the learned features within the different layers of a neural network can be hard. In this talk we will visually showcase the results of transforming the outputs of several layers of a state of the art imagenet architecture with UMAP.

Matthias Anderer, Geschäftsführer, Matthias Anderer GmbH