MAGIC: Modular Auto-encoder for Generalisable Model Inversion with Bias Corrections

📅 2024-05-29
🏛️ arXiv.org
📈 Citations: 1
Influential: 0
📄 PDF
🤖 AI Summary
Physical models in Earth observation often introduce systematic biases due to simplifications and incompleteness, severely compromising retrieval reliability. Existing approaches either neglect such biases or rely on interpretable auxiliary variables and complex regularization—both limiting practical applicability. This paper proposes an end-to-end, assumption-free framework that jointly performs geophysical inversion and bias correction. It embeds physics-based forward models (e.g., radiative transfer, volcanic deformation) into the decoder path of an autoencoder and cascades them with differentiable, learnable bias-correction layers. We introduce a modular physics-aware decoding architecture that explicitly decouples model priors from bias modeling—enabling cross-task generalization without pre-specifying bias forms. Evaluated on real-world remote sensing and geodetic tasks, our method achieves inversion accuracy comparable to or exceeding classical Bayesian and regression-based methods, while eliminating manual bias filtering.

Technology Category

Application Category

📝 Abstract
Scientists often model physical processes to understand the natural world and uncover the causation behind observations. Due to unavoidable simplification, discrepancies often arise between model predictions and actual observations, in the form of systematic biases, whose impact varies with model completeness. Classical model inversion methods such as Bayesian inference or regressive neural networks tend either to overlook biases or make assumptions about their nature during data preprocessing, potentially leading to implausible results. Inspired by recent work in inverse graphics, we replace the decoder stage of a standard autoencoder with a physical model followed by a bias-correction layer. This generalisable approach simultaneously inverts the model and corrects its biases in an end-to-end manner without making strong assumptions about the nature of the biases. We demonstrate the effectiveness of our approach using two physical models from disparate domains: a complex radiative transfer model from remote sensing; and a volcanic deformation model from geodesy. Our method matches or surpasses results from classical approaches without requiring biases to be explicitly filtered out, suggesting an effective pathway for understanding the causation of various physical processes.
Problem

Research questions and friction points this paper is trying to address.

Augments incomplete physical models with low-rank residuals
Improves interpretability and accuracy in Earth Observation inversions
Reduces prediction errors in forest and volcanic deformation analyses
Innovation

Methods, ideas, or system contributions that make the work stand out.

Augments incomplete physical models with learnable low-rank residual
Improves flexibility while staying close to governing physics
Enhances interpretability and accuracy in Earth Observation inversion
🔎 Similar Papers
No similar papers found.