The Transform-and-Perform framework: Explainable deep learning beyond classification

• DOI: 10.36227/techrxiv.21346425.v1
• Publication Date: 2022-10-26T02:06:22Z
• Authors: Vidya Prasad; Ruud van Sloun; Stef van den Elzen; Anna Vilanova; Nicola Pezzotti
• Keywords: Structuring; Popularity; Black box

<p>In recent years, visual analytics (VA) has shown promise in alleviating the challenges of interpreting black-box deep learning (DL) models. While focus VA for explainable DL been mainly on classification problems, is gaining popularity high-dimensional-to-high-dimensional (H-H) problems such as image-to-image translation. In contrast to classification, H-H have no explicit instance groups or classes study. Each output continuous, high dimensional, and changes an unknown non-linear manner with input. These relations between input, model necessitate user analyze them conjunction, leveraging symmetries them. Since tasks do not exhibit some these challenges, most existing systems frameworks allow limited control components required models beyond classification. Hence, we identify need present a unified conceptual framework, Transform-and-Perform framework (T&P), facilitate design analysis focusing problems. T&P provides guidelines structure workflows strategies new systems, understand ones uncover potential gaps improvements. The goal aid creation effective that support structuring understanding identifying actionable insights We highlight growing like real-world translation application. also illustrate how effectively supports systems.</p>