Interactive Dimensionality Reduction for Comparative Analysis

Fujiwara, Takanori; Wei, Xinhai; Zhao, Jian; Ma, Kwan-Liu

doi:10.48550/arxiv.2106.15481

Cited by 1 publication

(1 citation statement)

References 75 publications

(109 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For example, visual augmentations to highlight distortions in DR projections have included superimposed heat maps [2,39], lines whose lengths indicate the degree of projection error [41], and animations between projection axes [12]. In addition, a few systems have incorporated multiple projections, either for interactive parameter selection [13,34] or to compare proximity relationships in different variants [10]. In this work, we attempt to extend some of these techniques to the problems specific to embeddings in ML, including dataset scale and the need to compare more than two spaces.…”

Section: Dimensionality Reduction Visualizationmentioning

confidence: 99%

Emblaze: Illuminating Machine Learning Representations through Interactive Comparison of Embedding Spaces

Sivaraman

Perer

2022

27th International Conference on Intelligent User Interfaces

View full text Add to dashboard Cite

Fig. 1. Interactive comparison of embeddings for 50,000 words from three different corpora (a), using Emblaze within a Jupyter notebook. Given a selection, in this case a group of words about winemaking (b), the Star Trail visualization (c) highlights points near the group in the high-dimensional space whose neighborhoods change significantly. The plot can be manually interpolated between the two embedding spaces using a slider (d). In the sidebar, the neighborhood comparisons for the current selection (e) show a greater emphasis on recreational wine-related activities in Twitter data, such as "tastings" and "Sonoma. " The sidebar can alternatively display Suggested Selections relevant to the current visualization state (f), including clusters of wine varieties and other beverages.Modern machine learning techniques commonly rely on complex, high-dimensional embedding representations to capture underlying structure in the data and improve performance. In order to characterize model flaws and choose a desirable representation, model builders often need to compare across multiple embedding spaces, a challenging analytical task supported by few existing tools. We first interviewed nine embedding experts in a variety of fields to characterize the diverse challenges they face and techniques they use when analyzing embedding spaces. Informed by these perspectives, we developed a novel system called Emblaze that integrates embedding space comparison within a computational notebook environment. Emblaze uses an animated, interactive scatter plot with a novel Star Trail augmentation to enable visual comparison. It also employs novel neighborhood analysis and clustering procedures to dynamically suggest groups of points with interesting changes between spaces. Through a series of case studies with ML experts, we demonstrate how interactive comparison with Emblaze can help gain new insights into embedding space structure. CCS Concepts: • Human-centered computing → Visualization systems and tools; • Computing methodologies → Learning latent representations.

show abstract