Workshop: ISAV26: In Situ AI, Analysis and Visualization
Authors: Yuta Nakasaki (Kobe University, RIKEN Center for Computational Science (R-CCS)); Kinari Matsuzaki (Kobe University); Naohisa Sakamoto (Kobe University, RIKEN Center for Computational Science (R-CCS)); and Jorji Nonaka (RIKEN Center for Computational Science (R-CCS))
Abstract: Image-based in-situ visualization has become a mature approach that not only helps mitigate notorious I/O bottlenecks, but also facilitates exploratory visual analysis by enabling interactive exploration in the spatiotemporal domain. This approach generates a large collection of visual representations from different viewpoints, and in this work, we extend this approach to surrogate modeling, using the generated visualization images as a common representation that bridges human-centered interactive visual analysis and machine learning. During in-situ visualization, metadata required by existing information-entropy-based smart visualization techniques are also generated. Unlike previous smart in-situ visualization approaches, which try to output only the set of most information-rich rendering images, the proposed framework also outputs the complete set of rendered images for offline surrogate modeling. Our initial results demonstrate the feasibility of training surrogate models directly from the image-based representation generated for visual analysis, enabling learned approximation without requiring the original simulation data.
Back to ISAV26: In Situ AI, Analysis and Visualization Archive Listing
Back to Full Workshop Archive Listing