Alex Bäuerle, Mitglied der Forschungsgruppe Visual Computing, stellt sein Promotionsvorhabens unter dem Titel Visualization Interfaces for Different Stakeholders in the ML Pipeline vor.
Abstract: Artificial intelligence (AI) and the use of neural networks (NN) have become omnipresent in recent years. Neural networks model complex mathematical functions that can be based on billions, or even trillions, of parameters. At the same time, neural network s make decisions that can deeply impact our lives. Therefore, understanding, testing, and interpreting these networks and their decisions is an important area of research. While there exist introspection techniques that support such understanding, testing, and interpretation, their adoption for diagnosing systems and explaining decisions is difficult. Possible problems with the adoption of introspection tools are that current introspection tools are not easily implemented in one's framework, do not work wel l in combination to create more meaningful analysis, and are difficult to interpret.
Through the integration of existing and novel introspection tools into visualization interfaces, extensive analysis, actionable insights, and effective diagnosis are made widely available. These visualization interfaces can be integrated into existing development workflows, are designed to support flexible analysis needs, and thus make the interpretation of findings easier. I present visualization interfaces in three different areas, namely quality assurance, AI education, and communication, including a visualization approach for correcting mislabeled training data, an educational environment for recurrent neural networks (RNNs), and an interface for automatic network figure generation to communicate network architectures. Finally, to unify the diverse landscape of AI visualizations, I also present a framework for composing such interactive machine learning (ML) interfaces.
