A Deep Generative Model for Graph Layout

摘要: Different layouts can characterize different aspects of the same graph. Finding a “good” layout graph is thus an important task for visualization. In practice, users often visualize in multiple by using methods and varying parameter settings until they find that best suits purpose However, this trial-and-error process haphazard time-consuming. To provide with intuitive way to navigate design space, we present technique systematically diverse deep generative models. We encoder-decoder architecture learn model from collection example layouts, where encoder represents training examples latent space decoder produces space. particular, train construct two-dimensional easily explore generate various layouts. demonstrate our approach through quantitative qualitative evaluations generated The results show capable learning generalizing abstract concepts not just memorizing examples. summary, paper presents fundamentally new visualization machine learns without manually-defined heuristics.