Visualizing the Hidden Activity of Artificial Neural Networks

Research output: Journal contributionsJournal articlesResearchpeer-review

Authors

In machine learning, pattern classification assigns high-dimensional vectors (observations) to classes based on generalization from examples. Artificial neural networks currently achieve state-of-the-art results in this task. Although such networks are typically used as black-boxes, they are also widely believed to learn (high-dimensional) higher-level representations of the original observations. In this paper, we propose using dimensionality reduction for two tasks: visualizing the relationships between learned representations of observations, and visualizing the relationships between artificial neurons. Through experiments conducted in three traditional image classification benchmark datasets, we show how visualization can provide highly valuable feedback for network designers. For instance, our discoveries in one of these datasets (SVHN) include the presence of interpretable clusters of learned representations, and the partitioning of artificial neurons into groups with apparently related discriminative roles.

Original languageEnglish
Article number7539329
JournalIEEE Transactions on Visualization and Computer Graphics
Volume23
Issue number1
Pages (from-to)101-110
Number of pages10
ISSN1077-2626
DOIs
Publication statusPublished - 01.2017
Externally publishedYes

Recently viewed

Activities

  1. Investigating learner pragmatics at the discourse level: Challenges, complexities and insights
  2. International Conference on Mathematical Models & Computational Techniques in Science & Engineering - MMCTSE2020 
  3. Coherent sets in nonautonomous dynamics
  4. Acceptance and Feasibility of a mobile application for panic with and without agoraphobia
  5. Towards a Research Program in Design, Inovation and Aesthetics
  6. Coauthoring an interorganizational collaboration: Exploring multi-voicedness and introducing spatiotemporal orientations
  7. Workshop „Meta-Image Day 2012”
  8. Robots versus Machines
  9. An Evidence-based Approach to the Assessment of Public Participation in Environmental Governance: A conceptual and methodological overview of the ‘EDGE’ project
  10. Kunstuniversität Linz
  11. HyperKult XXI. 2012
  12. Dynamics Days Europe 2018
  13. Rudimentary theory of entrepreneurship
  14. How teachers use digital data: a systematic review
  15. 2nd International Conference on Mathematics and Computers in Science and Engineering - MACISE 2020
  16. Linking Teaching and Learning Formats with Student Development of Key Sustainability Competencies
  17. School Innovation through Knowledge Flows: Does Open Innovation Make the Difference?
  18. Do we need a new paradigm for mastering existing and future challenges of the urban water cycle
  19. Corporate Responsibility Evolution Models: Concepts, Evidence and Implications
  20. Ocean eddies and the polar vortex: coherence in complex systems
  21. Sym­po­si­um "Degrees of Freedom. Art Programs at Universities"
  22. Lodz University of Technology
  23. Art and Sustainability: Aesthetics of Complexity
  24. Conceptualising Animal Resistance: Questions and Approaches
  25. Workshop: Schreibtypengerecht Schreiben - 2018
  26. Journal of Pragmatics (Zeitschrift)
  27. Reciprocity and Democracy
  28. 20th Annual European Symposium on Algorithms - ESA 2012
  29. Where tasks, technology, and textbooks meet: Intelligent tutoring systems on the task-based language teacher's horizon (SLTED, Universität Wien)
  30. Peter Lang Verlag (Verlag)