Table 2 Summary of surrogate modeling strategies evaluated in this study

From: Accurate and uncertainty-aware multi-task prediction of HEA properties using prior-guided deep Gaussian processes

Model name

Uses prior

Tasks included

Description

HDGP P-All

Yes

Main + Auxiliary

DGP with priors derived from an encoder-decoder model trained on main tasks.

HDGP P-Main

Yes

Main only

DGP with priors from an encoder-decoder model trained only on main tasks.

HDGP NP-All

No

Main + Auxiliary

DGP without using prior knowledge.

HDGP NP-Main

No

Main only

DGP without priors or auxiliary properties.

GP (no corr. kernel)

No

Individual

Conventional GP trained independently for each property. Does not model inter-property correlations.

XGBoost

No

Per task

Gradient-boosted decision tree model applied separately to each property (classical regression baseline).

Encoder-Decoder (regularized dense network)

No

Main only

Multi-target regression model based on decoding nonlinearities in data (also used as a prior in HDGP configuration).

Back to article page