Omrani (2026) Modeling Interacting Convection Regimes with PINNs: Improving Vanilla Architecture Performance with Minimal Supervision- Trained Models

Identification

• Journal: Mendeley Data
• Year: 2026
• Date: 2026-01-05
• Authors: Mostafa Omrani
• DOI: 10.17632/stjnx2d98x

Research Groups

• KN Toosi University of Technology
• Amirkabir University of Technology

Short Summary

This paper models interacting convection regimes using Physics-Informed Neural Networks (PINNs), demonstrating improved performance of vanilla PINN architectures with minimal supervision.

Objective

• To model interacting convection regimes using Physics-Informed Neural Networks (PINNs) and enhance the performance of vanilla PINN architectures through minimal supervision.

Study Configuration

• Spatial Scale: Not explicitly detailed; likely a computational domain for fluid dynamics.
• Temporal Scale: Not explicitly detailed; refers to the duration of computational simulations.

Methodology and Data

• Models used: Physics-Informed Neural Networks (PINNs), specifically focusing on "vanilla architecture" improvements.
• Data sources: Not explicitly detailed, but implies limited supervised data for training the PINNs.

Main Results

• The study successfully demonstrates an improvement in the performance of vanilla Physics-Informed Neural Network (PINN) architectures when applied to interacting convection regimes, achieved with minimal supervisory data during training.

Contributions

• Introduces an approach to enhance the performance of vanilla PINN architectures for modeling complex interacting convection regimes, specifically highlighting the effectiveness of minimal supervision in achieving these improvements.

Funding

• Not specified in the provided text.

Citation

@article{Omrani2026Modeling,
  author = {Omrani, Mostafa},
  title = {Modeling Interacting Convection Regimes with PINNs: Improving Vanilla Architecture Performance with Minimal Supervision- Trained Models},
  journal = {Mendeley Data},
  year = {2026},
  doi = {10.17632/stjnx2d98x},
  url = {https://doi.org/10.17632/stjnx2d98x}
}