1 of 1

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

Scientific Achievement

An importance-driven method for generating domain-specific latent representations by autoencoders with reduced storage cost.

Significance and Impact

Scientists can define arbitrary importance criteria to obtain tailored latent representations for complex post-hoc analysis.

Technical Approach

Represent scientific interests with spatial importance maps
Generate domain interest guided latent representations through an autoencoder
A quantization and lossless data reduction component in the latent space to reduce data size

Jingyi Shen, Haoyu Li, Jiayi Xu, Ayan Biswas, and Han-Wei Shen: IDLat: An Importance-Driven Latent Generation Method for Scientific Data, IEEE Transactions on Visualization and Computer Graphics (Proc. IEEE VIS 2022)

Top: more regions become unimportant

Bottom: difference map spreads out more

Comparisons of Bruckner and Moller’s (left) and IDLat’s isosurfaces selection (right)

Baseline and IDLat’s reconstruction with value-based Importance maps

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 work, we develop 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. We represent data of interest by spatial importance maps and utilize the location-wise importance information to guide latent generation. With a trained model, scientists can flexibly define various importance criteria and obtain different latent representations. We further reduced the latent size by a lossless entropy encoding algorithm trained together with the autoencoder, improving the storage and memory efficiency.

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. Through quantitative and qualitative evaluations, we validate the effectiveness of our importance-driven latent generation method in representing data under domain interest control.