Reproduce “Understanding Bias in Perceiving Dimensionality Reduction Projections” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Seoyoung Doh, Hyeon Jeon, Sungbok Shin, Ghulam Jilani Quadri, Nam Wook Kim, Jinwook Seo
Researchers quantified how practitioners favor visually appealing dimensionality reduction projections over structurally faithful ones, even when objective metrics suggest otherwise. A user study revealed this bias is amplified by color encoding and linked to visual features like class separability, impacting the reliability of visual analytics.
Key insights
- Visual interestingness strongly influences practitioners' analytical preference for DR projections, often overriding contradictory objective faithfulness scores.
- The bias towards visual interestingness is significantly amplified when class labels are color-encoded (polychrome projections), as clear class separability becomes a dominant visual driver.
- This bias is rooted in cognitive processes, where salient visual cues activate fast, intuitive System 1 thinking, while evaluating numerical faithfulness scores requires slower, analytical System 2 engagement.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.