Physics-Informed Neural Networks: Advancements and Applications

Overview

Scientific computing is undergoing a paradigm shift with the integration of machine learning techniques that embed physical principles. Among these, physics-informed neural networks (PINNs) have emerged as a powerful approach for solving complex differential equations and modeling real-world systems by fusing physics-based knowledge with data-driven learning. Unlike purely data-centric methods, PINNs provide a framework that respects governing physical laws, leading to improved accuracy, generalization, and efficiency. The increasing relevance of PINNs across engineering, applied mathematics, and physical sciences positions them as a cornerstone for advancing computational intelligence in scientific discovery and real-world problem-solving.

Objective

This special session aims to establish a dedicated platform for the PINN research community, bringing together experts from machine learning, applied mathematics, physics, and engineering. The session will serve as a catalyst for cross-disciplinary dialogue, knowledge transfer, and collaboration to advance theoretical foundations, algorithmic strategies, and domain-specific applications of PINNs. By doing so, it will nurture both foundational research and real-world implementations, strengthening the role of PINNs as a unifying methodology in scientific machine learning. This session continues the PINN-focused special session at WCCI 2024. This special session is intended to be a part of WCCI 2026.

Key Topics

This session delves into key topics shaping the forefront of PINN research, including but not limited to:

• Physics-informed operator learning strategies: neural operators, learning solution operators, physics-guided training for sample efficiency.

• Foundations and theory of PINNs – mathematical formulations, well-posedness, convergence guarantees.

• Multiscale modeling – bridging microscale and macroscale phenomena via PINNs.

• Adaptive learning strategies – curriculum learning, residual-based adaptivity, and multi-fidelity training.

• Uncertainty quantification – probabilistic PINNs, Bayesian extensions, and error bounds.

• Transfer learning across physical domains – leveraging knowledge across engineering and scientific applications.

• Self-supervised and unsupervised strategies – PINNs without extensive labeled data.

• Active learning and sparse sampling – efficient data acquisition guided by physics priors.

• Hybrid modeling – combining PINNs with physics-based solvers or data-driven architectures.

• Sparse and efficient PINN variants – addressing scalability and computational costs.

• Generative models – for simulations, inverse problems, and design optimization.

• Novel domain applications – civil engineering, fluid mechanics, biomedical systems, climate modeling, and intelligent infrastructures.

Important Dates

• Paper Submission Deadline: January 31, 2026 (AoE)

• Final Decision Notification to Authors: March 15, 2026 (AoE)

• Camera-Ready Submission Deadline: April 15, 2026 (AoE)

Organisers