Deep Learning-Based Feature-Aware Data Modeling for Complex Physics Simulations
This addresses the need for feature-driven methods in in situ data analysis for exascale computing, though it appears incremental as it builds on existing autoencoder techniques.
The paper tackled the problem of feature-aware data reduction for in situ physics simulations by proposing a deep learning-based workflow using a Residual Autoencoder with RRDB, achieving compression from 2.1 MB to 66 KB per 3D volume timestep.
Data modeling and reduction for in situ is important. Feature-driven methods for in situ data analysis and reduction are a priority for future exascale machines as there are currently very few such methods. We investigate a deep-learning based workflow that targets in situ data processing using autoencoders. We propose a Residual Autoencoder integrated Residual in Residual Dense Block (RRDB) to obtain better performance. Our proposed framework compressed our test data into 66 KB from 2.1 MB per 3D volume timestep.