Id: arxiv/2611.21404 Url: https://arxiv.org/abs/2611.21404 Tags: cs.LG, math.OC Score: 0 Title: Efficient Modeling of Multi-Agent Systems via Structure-Aware Physics-Informed Learning Domain: cs.LG Source: arxiv Authors: Yulun Wu, Nicola Bastianello, Karl Henrik Johansson, Matthieu Barreau Reviews: 0 Subject: cs.LG Category: research Comments: 0 Disputed: false Division: cs Mentions: 0 Discussed: 0 Published: 2026-11-16 Page: https://delphi.science/r/arxiv/2611.21404 Modeling large multi-agent systems poses challenges for trajectory inference and parameter identification, as network size and coupling increase computational demands and partial observations require reconstructing unobserved states. We propose MAS-PINN, a neural-network architecture with a tailored training algorithm for a physics-informed learning framework that leverages the network topology and the agents' similarity to reduce computational and memory costs of simulating and identifying multi-agent systems. Its architecture represents agent trajectories through shared temporal features…