Adversarial Disentanglement by Backpropagation with Physics-Informed Variational Autoencoder

2025-06-16

AI Summary

To address inaccurate inference and prediction caused by partial observability of physical systems and interference from multiple confounding factors, this paper proposes a Physics-Informed Variational Autoencoder (PI-VAE). Methodologically, PI-VAE employs a modular latent space architecture that explicitly disentangles physics-constrained variables—governed by domain-specific priors—from data-driven variables capturing unknown disturbances. To preserve interpretability of the physics variables, an adversarial backpropagation constraint is introduced to prevent their entanglement with data components; weakly supervised disentanglement is further enforced via class- or domain-level observable labels. Evaluated on multiple synthetic engineering structural scenarios, PI-VAE achieves substantial improvements in reconstruction accuracy under low-data and noisy conditions (average gain of 23.6%) and demonstrates superior cross-operating-condition generalization. Notably, it is the first framework to enable interpretable and separable modeling of underlying physical mechanisms and unknown confounders within the latent space.

Technology Category

• Reasoning under Uncertainty: Probabilistic Inference
• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Low Level & Physics-based Vision

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semantics

Abstract

Inference and prediction under partial knowledge of a physical system is challenging, particularly when multiple confounding sources influence the measured response. Explicitly accounting for these influences in physics-based models is often infeasible due to epistemic uncertainty, cost, or time constraints, resulting in models that fail to accurately describe the behavior of the system. On the other hand, data-driven machine learning models such as variational autoencoders are not guaranteed to identify a parsimonious representation. As a result, they can suffer from poor generalization performance and reconstruction accuracy in the regime of limited and noisy data. We propose a physics-informed variational autoencoder architecture that combines the interpretability of physics-based models with the flexibility of data-driven models. To promote disentanglement of the known physics and confounding influences, the latent space is partitioned into physically meaningful variables that parametrize a physics-based model, and data-driven variables that capture variability in the domain and class of the physical system. The encoder is coupled with a decoder that integrates physics-based and data-driven components, and constrained by an adversarial training objective that prevents the data-driven components from overriding the known physics, ensuring that the physics-grounded latent variables remain interpretable. We demonstrate that the model is able to disentangle features of the input signal and separate the known physics from confounding influences using supervision in the form of class and domain observables. The model is evaluated on a series of synthetic case studies relevant to engineering structures, demonstrating the feasibility of the proposed approach.

Problem

• Disentangling known physics from confounding influences in data
• Improving generalization with limited noisy data using physics-informed models
• Ensuring interpretability of latent variables in hybrid physics-data models

Innovation

• Physics-informed variational autoencoder combines interpretability and flexibility
• Adversarial training ensures interpretable physics-grounded latent variables
• Latent space partitioned into physics-based and data-driven variables