Intelligent visualization and visual analytics
Tao Jun,Zhang Yu,Chen Qing,Liu Can,Chen Siming,Yuan Xiaoru(Key Laboratory of Machine Perception (Ministry of Education), and School of Intelligence Science and Technology, Peking University)
Visualization and visual analytics emerge as a common practice to analyze and understand data in the big data era. Visualization leverages visual encoding to transfer data into visual format, and visual analytics further involves user interaction through graphical interface to build a closed-cycle between users and data. These techniques allow users to build sophisticated data analysis workflow interactively and intuitively. However, the explosive growth of data raises great challenges to visual encoding and interactions: straightforward visual encoding becomes insufficient to represent the rich information of data in the limited screen-space, and the trail-and-error interaction is unaffordable to verify various hypotheses and explore all possibilities. To tackle these challenges, artificial intelligence is introduced in visualization approaches to optimize the information transfer among data, visualization, and users. The recent advancement in machine intelligence provides unprecedented power to represent data, visual representations, and user intention in a unified space and approximate complex connections among them. This may enhance the performance of conventional visualization approaches, and, more importantly, pave pathways to endless research possibilities beyond traditional visualization tasks. For example, in visualization creation, machine intelligence can capture important data features and user intentions through learning, which facilitates auto-creation of visualization that meets specific design requirements. This reduces the expertise and effort required to generate desired visualization from data. In scientific visualization, intelligent approaches can effectively learn from a large amount of ensemble members and produce visualization of unseen members. This enables real-time exploration of ensemble data without prohibitive simulation. In interactive techniques, machine learning can enrich interaction modalities to reduce learning and usage effort. One example is the natural language interface, which allows users to interact with visualization systems using natural language. In visual analytic systems, machine intelligence may learn users’ exploration behavior and suggest likely interaction operations. This reduces the trial-and-error effort and improve the interaction efficiency. This paper surveys the AI-based visualization techniques and discusses the trends and research opportunities. Particularly, we explore the role of AI in the key steps of the visualization pipeline: data representation, visualization creation, interactive exploration, and visual analytics. For data representation, we discuss how to leverage the power of AI for big data management and representation and support astonishing rendering. For visualization creation, we discuss how AI may facilitate effective transformation from data to its visual form preserving the key structures and highlighting interesting facts. For interactive exploration, we discuss how AI may be used to interpret user intention behind various kinds of interaction and how visualization systems response to the intention. For visual analytics, we focus on how AI may support the downstream analysis tasks. Additionally, we introduce the application of intelligent visualization and visual analytics with graphical data as an example. Finally, we will discuss the recent trends in intelligent visualization approaches and points out future research directions and application scenarios.