Concept Clusters

Information

• Publication Type: Student Project
• Workgroup(s)/Project(s):
  • Visualization Group
• Date: 2023
• Date (Start): March 2022
• Date (End): March 2023
• Matrikelnummer: 0925239
• First Supervisor: Manuela Waldner

Abstract

The Motivation for this work is to explore methods for visualizing high-dimensional datasets, with a specific focus on understanding the inner workings of Deep Neural Networks (DNNs). DNNs are a type of Artificial Neural Network (ANN) that consist of an input layer, an output layer, and multiple hidden layers. They are able to make predictions or decisions based on unseen observations, but the way that they arrive at these decisions can often be opaque, or difficult to interpret. By exploring the similarities of feature vectors in the latent space created by the network, it is possible to gain insight into how the network organizes and classifies observations with similar properties that are also interpretable to the human eye.

The goal of this work is to develop a prototype that enables interactive exploration of data clusterings in high-dimensional feature spaces, instead of the traditional twodimensional visual representation.

Additional Files and Images

Weblinks

No further information available.

BibTeX

@studentproject{wolf-2023-cc,
  title =      "Concept Clusters",
  author =     "Dominik Wolf",
  year =       "2023",
  abstract =   "The Motivation for this work is to explore methods for
               visualizing high-dimensional datasets, with a specific focus
               on understanding the inner workings of Deep Neural Networks
               (DNNs). DNNs are a type of Artificial Neural Network (ANN)
               that consist of an input layer, an output layer, and
               multiple hidden layers. They are able to make predictions or
               decisions based on unseen observations, but the way that
               they arrive at these decisions can often be opaque, or
               difficult to interpret. By exploring the similarities of
               feature vectors in the latent space created by the network,
               it is possible to gain insight into how the network
               organizes and classifies observations with similar
               properties that are also interpretable to the human eye. 
               The goal of this work is to develop a prototype that enables
               interactive exploration of data clusterings in
               high-dimensional feature spaces, instead of the traditional
               twodimensional visual representation.",
  month =      mar,
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2023/wolf-2023-cc/",
}