Interactive Volume Visualization of Large-scale Scientific Data Modeled by Functional Approximation

Volume visualization is widely employed to reveal insightful complex patterns of scientific datasets across various domains. However, scientific datasets are often characterized by their substantial size and multidimensional nature, resulting in complex structures with diverse scales and posing challenges in efficiently producing high-quality volume rendering outcomes. Although functional approximation, as a continuous representation, provides features of higher-order approximation, compact representation, and better rendering quality for scientific data, its slow querying latency becomes the main factor affecting the system’s responsiveness.

In this talk, I will discuss my three research works for improving the overall input latency of visualizing large-scale scientific data modeled by function approximation. The first work leverages the massively distributed computing power to develop a scalable interactive visualization pipeline for rendering large-scale data modeled by functional approximation. The second work provides a GPU-accelerated rendering framework on a single computing node leveraging out-of-core methods and multi-resolution. My last work further improves the responsiveness of typical volume visualization systems by considering user exploratory behavior through deep learning-based prefetching.

Bio: Jianxin (Jason) Sunis a research scientist in the School of Computing at the University of Nebraska-Lincoln. His research concentrates on large-scale scientific data modeling, analysis, and visualization. Jason received his Ph.D. in Computer Science from the University of Nebraska-Lincoln in May 2024.