arXiv:2606.19853v1 Announce Type: new Abstract: We introduce SEA-PINN, a novel architecture that incorporates a Squeeze-Excitation-like attention mechanism into physics-informed neural networks to dynamically recalibrate the importance of neurons across layers. A key feature of SEA-PINN is its highly stable initialization. On 17 out of 20 benchmark problems, SEA-PINN exhibit nearly negligible variance and significantly reduced initial loss, establishing a quasi-deterministic and favorable starting point for optimization. Notably, without employing Fourier feature embeddings or periodic activat
The continuous evolution of AI demands more stable and efficient training methods, and attention mechanisms are a current area of intense research.
Improved stability and reduced initial loss in PINNs can accelerate scientific discovery and engineering design processes across multiple domains.
The development of SEA-PINN offers a more robust and predictable foundation for applying neural networks to solve complex physics problems.
- · AI researchers
- · Engineering industries
- · Scientific computing
More reliable and faster convergence for physics-informed neural networks.
Accelerated development cycles for new materials, drug discovery, and climate models due to improved simulation capabilities.
Enhanced AI systems begin to autonomously optimize complex physical processes, leading to breakthroughs in manufacturing or energy systems.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG