Abstract
With the growing computational power available for high-resolution ensemblesimulations in scientific fields such as cosmology and oceanology, storage andcomputational demands present significant challenges. Current surrogate modelsfall short in the flexibility of point- or region-based predictions as theentire field reconstruction is required for each parameter setting, hencehindering the efficiency of parameter space exploration. Limitations exist incapturing physical attribute distributions and pinpointing optimal parameterconfigurations. In this work, we propose Explorable INR, a novel implicitneural representation-based surrogate model, designed to facilitate explorationand allow point-based spatial queries without computing full-scale field data.In addition, to further address computational bottlenecks of spatialexploration, we utilize probabilistic affine forms (PAFs) for uncertaintypropagation through Explorable INR to obtain statistical summaries,facilitating various ensemble analysis and visualization tasks that areexpensive with existing models. Furthermore, we reformulate the parameterexploration problem as optimization tasks using gradient descent and KLdivergence minimization that ensures scalability. We demonstrate that theExplorable INR with the proposed approach for spatial and parameter explorationcan significantly reduce computation and memory costs while providing effectiveensemble analysis.