Deep‑Surrogate Framework Predicts Quality of Error‑Bounded Lossy Compression

Global: Deep Surrogate Framework Predicts Quality of Error-Bounded Lossy Compression

New Framework Targets Compression Quality Prediction

Researchers have introduced a deep‑surrogate framework called DeepCQ that predicts the quality of error‑bounded lossy compression for scientific data. The preprint describing the work was posted on arXiv in December 2025. The system seeks to lower the computational expense of assessing compression quality, a process that traditionally relies on intensive metric calculations. By delivering accurate predictions, the framework enables scientists to decide whether a given compression level meets their data‑quality requirements.

Generalizable Surrogate Modeling

According to the authors, the surrogate model is designed to be applicable across a variety of compressors, quality metrics, and input datasets. This generality is intended to reduce the need for bespoke prediction tools for each new scientific application.

Two‑Stage Architecture Improves Efficiency

The framework employs a novel two‑stage design that separates the computationally heavy feature‑extraction phase from a lightweight metrics‑prediction phase. Consequently, training can be performed efficiently, and inference remains modular and fast.

Adapting to Evolving Simulation Data

A mixture‑of‑experts component is incorporated to enhance robustness when predicting across simulation timesteps. The authors note that this design helps maintain accuracy even when training and test data exhibit significant variation.

Empirical Validation Across Scientific Domains

The approach was evaluated on four real‑world scientific applications. Reported prediction errors were generally under 10 % across most settings, indicating high predictive accuracy.

Performance Gains Over Prior Techniques

Compared with existing methods, the authors claim that DeepCQ delivers substantially lower prediction errors, thereby offering more reliable quality assessments for compressed data.

Potential Impact on Scientific Workflows

If adopted, the framework could allow researchers to reduce I/O and computational overhead during data analysis by selecting compression settings that meet predefined quality thresholds without exhaustive metric computation.

This report is based on information from arXiv, licensed under Academic Preprint / Open Access. Based on the abstract of the research paper. Full text available via ArXiv.