IDLat: An Importance-Driven Latent Generation Method for Scientific Data

JINGYI SHEN, Haoyu Li, Jiayi Xu, Ayan Biswas, Han-Wei Shen

View presentation:2022-10-20T14:48:00ZGMT-0600Change your timezone on the schedule page
2022-10-20T14:48:00Z
Exemplar figure, described by caption below
We present an importance-driven latent generation method (IDLat) based on an autoencoder model which tightly relates latent representations to specific data of interest, such as salient regions or features of interest. With a trained model, scientists can flexibly define various importance criteria and obtain different latent representations. We further reduce the latent size through a lossless entropy coding model. In addition, we develop a visual exploration tool for latent space analysis and demonstrate the efficiency of identifying and analyzing feature regions with importance-driven latent representations.

Prerecorded Talk

The live footage of the talk, including the Q&A, can be viewed on the session page, ML for VIS.

Fast forward
Abstract

Deep learning based latent representations have been widely used for numerous scientific visualization applications such as isosurface similarity analysis, volume rendering, flow field synthesis, and data reduction, just to name a few. However, existing latent representations are mostly generated from raw data in an unsupervised manner, which makes it difficult to incorporate domain interest to control the size of the latent representations and the quality of the reconstructed data. In this paper, we present a novel importance-driven latent representation to facilitate domain-interest-guided scientific data visualization and analysis. We utilize spatial importance maps to represent various scientific interests and take them as the input to a feature transformation network to guide latent generation. We further reduced the latent size by a lossless entropy encoding algorithm trained together with the autoencoder, improving the storage and memory efficiency. We qualitatively and quantitatively evaluate the effectiveness and efficiency of latent representations generated by our method with data from multiple scientific visualization applications.