Melody: Generating and Visualizing Machine Learning Model Summary to Understand Data and Classifiers Together

• DOI: 10.48550/arxiv.2007.10614
• Publication Date: 2020-01-01
• Authors: Gromit Yeuk-Yin Chan; Enrico Bertini; Luís Gustavo Nonato; Brian Barr; Claudio Silva
• Keywords: FOS: Computer and information sciences; Computer Science - Human-Computer Interaction; 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology; Human-Computer Interaction (cs.HC)

With the increasing sophistication of machine learning models, there are growing trends developing model explanation techniques that focus on only one instance (local explanation) to ensure faithfulness original model. While these provide accurate interpretability various data primitive (e.g., tabular, image, or text), a holistic Explainable Artificial Intelligence (XAI) experience also requires global and dataset enable sensemaking in different granularity. Thus, is vast potential synergizing visual analytics approaches. In this paper, we present MELODY, an interactive algorithm construct optimal overview behavior by summarizing local explanations using information theory. The result (i.e., summary) does not require additional restrictions primitives, knowledge from users. We design MELODY UI, system demonstrate how summary connects dots XAI tasks inspections. three usage scenarios regarding text classifications illustrate generalize data. Our experiments show our approaches: (1) provides better compared straightforward information-theoretic summarization (2) achieves significant speedup end-to-end modeling pipeline.