Cieran: Designing Sequential Colormaps via In-Situ Active Preference Learning

Paper Title

Cieran: Designing Sequential Colormaps via In-Situ Active Preference Learning

Paper Information

• Subject Area: Data Visualization, Sequential Colormap Design and Optimization
• Keywords: Visualization, Colormap, Design Optimization, Preference Learning, Data Analysis

Research Background and Problem

• Research Problem:

  • Analysts often need to select appropriate sequential colormaps when designing visualizations, a process that is both technical and aesthetic, while being constrained by the limitations of default options or predefined color libraries.
  • Custom colormap tools (e.g., Photoshop and CCC-Tool) require technical and design expertise, posing a barrier for ordinary users.
  • There is a lack of methods in academia to quantify the aesthetic value of sequential colormaps, and existing tools pay insufficient attention to user personalization needs.

• Research Importance:

  • Sequential colormaps are critical in data visualization, as the correct choice of colormap can effectively convey data patterns.
  • Improving colormap design efficiency is particularly important for data analysts, many of whom lack the expertise to manually create customized colormaps.

• Research Motivation and Related Work:

  • Existing tools like ColorBrewer and matplotlib provide expert-designed colormap options, but users must manually sift through them, which is time-consuming.
  • Related works like ColorCrafter and ColorMoves attempt to simplify colormap design but still require users to engage in complex operations.
  • Active preference learning has been used in image optimization and may be extendable to personalized colormap design.

Solution

• Proposed Solution:

  • Develop a Python toolkit named Cieran, integrated into Jupyter Notebooks, to quickly generate sequential colormaps based on dynamic user preferences.

• Innovative Aspects of the Solution:

  • For the first time, the sequential colormap design problem is treated as a path planning problem in the CIELAB color space.
  • A preference learning algorithm is used to establish an aesthetic utility model linked to user personalization needs.
  • The design process automatically sorts and creates colormaps while ensuring all outputs adhere to perceptual guidelines (e.g., linear order, smoothness, perceptual uniformity).

• Implementation Steps and Key Techniques:

  • Initialization: Users initiate Cieran by specifying seed colors on the target visualization.
  • Preference Learning: The system trains the model by presenting colormap pairs and collecting user preferences (e.g., which is better or no difference).
  • Result Generation: The model is used to rank expert-designed colormaps and create new colormaps that meet user preferences.
  • Key technical highlights include:
    • Intuitive quantification of colormap curves based on the CIELAB color space.
    • Dynamic collection of effective feedback using a preference learning algorithm.
    • Employing reinforcement learning to find efficient paths, avoiding inefficient design caused by color choices.

Research Outcomes

• Specific Outcomes:

  • Cieran effectively learns user preferences and recommends personalized, perceptually guided sequential colormaps within approximately two minutes.
  • The system was experimentally validated to perform well in both user preference ranking and the creation of new colormaps.

• Advantages Compared to Existing Solutions:

  • Compared to manual design and other tools, Cieran significantly reduces user design time and technical barriers while enhancing personalization and aesthetic satisfaction.
  • It can quickly create new colormaps based on user preferences and effectively rank hundreds of expert-designed colormaps.

• Experimental or Evaluation Results:

  • User studies show that Cieran can accurately rank colormaps, and the new colormaps it creates outperform lower-ranked expert-designed options in most cases.
  • The path planning method based on an optimized Q-learning algorithm performs excellently in generating colormap trajectories.

• Limitations and Future Directions:

  • Limitations:
    • The program relies on expert-designed color libraries, and the quantity and diversity of colormaps may limit design effectiveness.
    • Display device limitations may hinder the effective presentation of certain colormaps.
  • Future Directions:
    • Expand the color library to support more seed colors and enhance design flexibility.
    • Explore dynamically combining accessibility (e.g., colorblind contrast) with user preferences.
    • Provide context-driven design support, such as adjusting hue variation ranges or automatically matching domain-specific conventions based on the target task.

Conclusion

Cieran provides data analysts with an efficient tool for optimizing sequential colormap design through active preference learning, meeting aesthetic and perceptual needs in specific scenarios. This work not only introduces new technical approaches for design automation tools but also highlights the potential for addressing diverse user needs in visualization design.