可视化的设计空间庞大复杂，参数空间选择困难，用户的感知空间难以直接表达，本项目旨在针对上述可视化中设计空间、参数空间、感知空间的难点，研究由机器学习驱动的可视化新流程取代需要针对特定数据和任务设立不同可视化设计的模式，通过学习的方式建立隐式的模型。将可视化本身以及相关的参数设置、设计空间、用户感知等信息，通过选择合适的机器学习模型与算法之后进行学习、训练，直接合成期望的可视化参数、设置、甚至可视化结果，从而提供更智能的可视化。这项工作将引入和当前一般可视化不同的可视化设计流程，将可视化设计范例从传统的解析式驱动方法转变为目标导向方式。

In a typical visualization, the design space could be very huge, and the parameter space may have too many parameters to tune. It is also difficult to model the users' perception. To tackle the challenges of design space, parameter space and perception space, the goal of this project is to replace previously difficult elements of the visualization pipeline, which are normally specifically designed for dedicated data and tasks that have defined analytical formulas, with more powerful and generic machine learning approaches. These machine learning approaches will either take example results from existing algorithms or results generated by human users as input. With proper model training based on the input data, and example results with designated learning algorithms, it will be possible to directly synthesize the desired visualization parameters, settings or even results. More intelligent visualization can be provided through our approaches. The approaches proposed in this paper are fundamentally different from those in current visualization practice. In our approach, we propose to shift the visualization paradigm from one that is parameter or analytical form-driven to one that is goal-oriented.