Introduction
This special section of the IEEE Transactions on Visualization and Computer Graphics (IEEE TVCG) presents five of the most highly rated papers from the 2023 IEEE Pacific Visualization Symposium (IEEE PacificVis), held in Seoul, Korea. Sponsored by the IEEE Visualization and Graphics Technical Committee (VGTC), PacificVis has evolved into a premier international forum that fosters exchange between researchers and practitioners. While it serves as a vital hub for the Asia-Pacific region, its reach is truly global, attracting contributions that push the boundaries of visual analytics, privacy-preserving techniques, and explainable artificial intelligence (XAI).
Navigating Complex Data Landscapes
The papers selected for this special section address critical gaps in how we process, protect, and interpret complex data. From the challenges of data silos in social science to the "black-box" nature of modern AI, these works offer innovative methodological solutions.
1. SDRQuerier: Harmonizing Survey Data
Cross-national public opinion surveys are often fragmented across different databases, making it difficult for researchers to identify relational patterns. SDRQuerier introduces a visual querying framework that allows users to navigate harmonized survey databases using research questions or keywords. By disclosing underlying relational patterns, it bridges the gap between disparate data sources and meaningful sociological insights.
2. Federated Visualization: Privacy by Design
As data privacy regulations tighten, the challenge of visualizing decentralized data without moving raw records becomes paramount. Inspired by "federated learning" in machine learning, this paper proposes a Federated Visualization architecture. It utilizes a shared global module to aggregate encrypted, transformed visual features from local data modules. This allows for global data exploration and querying while ensuring that individual privacy remains uncompromised.
3. LinSets.zip: Optimizing Visual Real Estate
Linear set diagrams are powerful tools for showing memberships, but they suffer from a "sparsity" problem as the number of sets increases, leading to unreadable displays. LinSets.zip addresses this through a compression technique powered by integer-linear programming (ILP) and heuristics. By mathematically re-arranging set memberships into a more space-efficient matrix, the system significantly reduces the footprint of the diagram without losing accuracy or task performance.
4. FraudAuditor: Unmasking Collusive Fraud
Detecting health insurance fraud is notoriously difficult due to the lack of labeled data and the sophisticated nature of collusive behavior. FraudAuditor is a three-stage visual analytics system that shifts the focus from individual records to "co-visit networks." By modeling the relationships between patients and using improved community detection algorithms, the system allows experts to identify and verify suspicious, collusive groups that traditional statistical methods might overlook.
5. Interpreting Vision Transformers (ViTs)
Vision Transformers have revolutionized computer vision, yet their internal decision-making processes—specifically their attention heads—remain opaque. This research utilizes pruning-based metrics and autoencoder-based learning to profile the spatial distribution of attention strengths. By identifying which heads are most influential and summarizing their patterns, the tool provides a clearer window into how these models process visual information.
Impact and Results
The efficacy of these systems has been validated through both quantitative and qualitative measures:
- Efficiency: LinSets.zip demonstrated that compressed diagrams perform on par with traditional layouts in terms of user accuracy and speed, despite their smaller size.
- Expert Validation: FraudAuditor and the ViT analysis tool were vetted by domain experts, confirming their utility in real-world healthcare and AI development scenarios.
- Architectural Innovation: The Federated Visualization model proves that visualization can adopt decentralized architectures to meet modern security standards.
Conclusion
The papers featured in this special section represent the vanguard of visualization research. By addressing scalability, privacy, and interpretability, these authors contribute to a more robust and transparent data ecosystem. As IEEE PacificVis continues to grow, it remains a critical bridge for sharing the latest advances between the Asia-Pacific region and the broader international research community.
Summary updated