The Transform-and-Perform Framework: Explainable Deep Learning Beyond Classification

Abstract

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

ProactiV: Studying Deep Learning Model Behavior Under Input Transformations

“…However, the input features or meta-data needed to determine data slices are often unavailable in image-to-image translation tasks. Further, studying trends beyond differential behavior is critical in image translation tasks where there is no clear distinction between right and wrong outputs [28].…”

“…It is hence challenging to study global effects under multiple input transforms. Moreover, prior work focuses on classification, object detection, and segmentation tasks [27], [28]. However, image translation tasks with high-dimensional regression outputs are hardly supported [28].…”

“…Nevertheless, prior work does not focus on providing a global behavioral overview under multiple input transformations. Additionally, prior work often focuses on classification or object detection tasks, while more generic and complex computer vision tasks are not explored [27], [28]. Specifically, DL models are generic parametric functions y = F θ (x), where inputs x are processed by the model F θ having parameters θ , to generate outputs y.…”

“…In classification, the output y is categorical, and its dimensionality m is in the order of tens. With newer DL developments such as image-to-image translation, outputs y have no explicit groups and are high-dimensional (m ≃ n) regression variables, increasing complexities [28].…”

State of the Art of Visual Analytics for eXplainable Deep Learning

“…Most VA systems target standard CNNs used in a classification context with few classes without necessarily using state‐of‐the‐art XDL methods and without exploring the most challenging and recent problems, as also noted by [PvSvdE*22]. We think systems should be more versatile by focusing on a deeper variety of DL model families and by using more recent XDL methods.…”