arXiv subjects

Thilo Strauss

Publications and source records attributed to Thilo Strauss.

12 recordsLinked to original sources

arXiv2025

Self-adaptive physics-informed neural network for forward and inverse problems in heterogeneous porous flow

We develop a self-adaptive physics-informed neural network (PINN) framework that reliably solves forward Darcy flow and performs accurate permeability inversion in heterogeneous porous media. In the forward setting, the PINN predicts velocity and pressure for discontinuous, piecewise-constant permeability; in the inverse setting, it identifies spatially varying permeability directly from indirect flow observations. Both models use a region-aware permeability parameterization with binary spatial masks, which preserves sharp permeability jumps and avoids the smoothing artifacts common in standard PINNs. To stabilize training, we introduce self-learned loss weights that automatically balance PDE residuals, boundary constraints, and data mismatch, eliminating manual tuning and improving robustness, particularly for inverse problems. An interleaved AdamW-L-BFGS optimization strategy further accelerates and stabilizes convergence. Numerical results demonstrate accurate forward surrogates and reliable inverse permeability recovery, establishing the method as an effective mesh-free solver and data-driven inversion tool for porous-media systems governed by partial differential equations.

physics.flu-dyn↗

arXiv2025

Preference-Based Gradient Estimation for ML-Guided Approximate Combinatorial Optimization

Combinatorial optimization (CO) problems arise across a broad spectrum of domains, including medicine, logistics, and manufacturing. While exact solutions are often computationally infeasible, many practical applications require high-quality solutions within a given time budget. To address this, we propose a learning-based approach that enhances existing non-learned approximation algorithms for CO. Specifically, we parameterize these approximation algorithms and train graph neural networks (GNNs) to predict parameter values that yield near-optimal solutions. Our method is trained end-to-end in a self-supervised fashion, using a novel gradient estimation scheme that treats the approximation algorithm as a black box. This approach combines the strengths of learning and traditional algorithms: the GNN learns from data to guide the algorithm toward better solutions, while the approximation algorithm ensures feasibility. We validate our method on two well-known combinatorial optimization problems: the travelling salesman problem (TSP) and the minimum k-cut problem. Our results demonstrate that the proposed approach is competitive with state-of-the-art learned CO solvers.

cs.LG↗

arXiv2024

MCMC-Net: Accelerating Markov Chain Monte Carlo with Neural Networks for Inverse Problems

In many computational problems, using the Markov Chain Monte Carlo (MCMC) can be prohibitively time-consuming. We propose MCMC-Net, a simple yet efficient way to accelerate MCMC via neural networks. The key idea of our approach is to substitute the true likelihood function of the MCMC method with a neural operator based surrogate. We extensively evaluate the accuracy and speedup of our method on three different PDE-based inverse problems where likelihood computations are computationally expensive, namely electrical impedance tomography, diffuse optical tomography, and quantitative photoacoustic tomography. MCMC-Net performs similar to the classical likelihood counterpart but with a significant speedup. We conjecture that the method can be applied to any problem with a sufficiently expensive likelihood function. We also analyze MCMC-Net in a theoretical setting for the different use cases. We prove a universal approximation theorem-type result to show that the proposed network can approximate the mapping resulting from forward model evaluations to a desired accuracy. Furthermore, we establish convergence of the surrogate posterior to the true posterior under Hellinger distance.

math.NA↗

arXiv2024

A DeepONet for inverting the Neumann-to-Dirichlet Operator in Electrical Impedance Tomography: An approximation theoretic perspective and numerical results

In this work, we consider the non-invasive medical imaging modality of Electrical Impedance Tomography (EIT), where the goal is to recover the conductivity in a medium from boundary current-to-voltage measurements, i.e., the Neumann-to-Dirichlet (N--t--D) operator. We formulate this inverse problem as an operator-learning task, where the aim is to approximate the implicitly defined map from N--t--D operators to admissible conductivities. To this end, we employ a Deep Operator Network (DeepONet) architecture, thereby extending operator learning beyond the classical function-to-function setting to the more challenging operator-to-function regime. We establish a universal approximation theorem that guarantees that such operator-to-function maps can be approximated arbitrarily well by DeepONets. Furthermore, we provide a computational implementation of our approach and compare it against the iteratively regularized Gauss--Newton (IRGN) method. Our results show that the proposed framework yields accurate and robust reconstructions, outperforms the baseline, and demonstrates strong generalization. To our knowledge, this is the first work that combines rigorous approximation-theoretic guarantees with DeepONet-based inversion for EIT, thereby opening a principled and interpretable pathway for use of DeepONets in such inverse problems.

cs.LG↗

arXiv2024

Simultaneous Estimation of Piecewise Constant Coefficients in Elliptic PDEs via Bayesian Level-Set Methods

In this article, we propose a non-parametric Bayesian level-set method for simultaneous reconstruction of two different piecewise constant coefficients in an elliptic partial differential equation. We show that the Bayesian formulation of the corresponding inverse problem is well-posed and that the posterior measure as a solution to the inverse problem satisfies a Lipschitz estimate with respect to the measured data in terms of Hellinger distance. We reduce the problem to a shape-reconstruction problem and use level-set priors for the parameters of interest. We demonstrate the efficacy of the proposed method using numerical simulations by performing reconstructions of the original phantom using two reconstruction methods. Posing the inverse problem in a Bayesian paradigm allows us to do statistical inference for the parameters of interest, whereby we are able to quantify the uncertainty in the reconstructions for both methods. This illustrates a key advantage of Bayesian methods over traditional algorithms.

stat.AP↗

arXiv2020

Learning Implicit Surface Light Fields

Implicit representations of 3D objects have recently achieved impressive results on learning-based 3D reconstruction tasks. While existing works use simple texture models to represent object appearance, photo-realistic image synthesis requires reasoning about the complex interplay of light, geometry and surface properties. In this work, we propose a novel implicit representation for capturing the visual appearance of an object in terms of its surface light field. In contrast to existing representations, our implicit model represents surface light fields in a continuous fashion and independent of the geometry. Moreover, we condition the surface light field with respect to the location and color of a small light source. Compared to traditional surface light field models, this allows us to manipulate the light source and relight the object using environment maps. We further demonstrate the capabilities of our model to predict the visual appearance of an unseen object from a single real RGB image and corresponding 3D shape information. As evidenced by our experiments, our model is able to infer rich visual appearance including shadows and specular reflections. Finally, we show that the proposed representation can be embedded into a variational auto-encoder for generating novel appearances that conform to the specified illumination conditions.

cs.CV↗

arXiv2020

Statistical Inversion Using Sparsity and Total Variation Prior And Monte Carlo Sampling Method For Diffuse Optical Tomography

In this paper, we formulate the reconstruction problem in diffuse optical tomography (DOT) in a statistical setting for determining the optical parameters, scattering and absorption, from boundary photon density measurements. A special kind of adaptive Metropolis algorithm for the reconstruction procedure using sparsity and total variation prior is presented. Finally, a simulation study of this technique with different regularization functions and its comparison to the deterministic Iteratively Regularized Gauss Newton method shows the effectiveness and stability of the method.

math.NA↗

arXiv2019

CANet: An Unsupervised Intrusion Detection System for High Dimensional CAN Bus Data

We propose a novel neural network architecture for detecting intrusions on the CAN bus. The Controller Area Network (CAN) is the standard communication method between the Electronic Control Units (ECUs) of automobiles. However, CAN lacks security mechanisms and it has recently been shown that it can be attacked remotely. Hence, it is desirable to monitor CAN traffic to detect intrusions. In order to detect both, known and unknown intrusion scenarios, we consider a novel unsupervised learning approach which we call CANet. To our knowledge, this is the first deep learning based intrusion detection system (IDS) that takes individual CAN messages with different IDs and evaluates them in the moment they are received. This is a significant advancement because messages with different IDs are typically sent at different times and with different frequencies. Our method is evaluated on real and synthetic CAN data. For reproducibility of the method, our synthetic data is publicly available. A comparison with previous machine learning based methods shows that CANet outperforms them by a significant margin.

cs.CR↗

arXiv2019

Texture Fields: Learning Texture Representations in Function Space

In recent years, substantial progress has been achieved in learning-based reconstruction of 3D objects. At the same time, generative models were proposed that can generate highly realistic images. However, despite this success in these closely related tasks, texture reconstruction of 3D objects has received little attention from the research community and state-of-the-art methods are either limited to comparably low resolution or constrained experimental setups. A major reason for these limitations is that common representations of texture are inefficient or hard to interface for modern deep learning techniques. In this paper, we propose Texture Fields, a novel texture representation which is based on regressing a continuous 3D function parameterized with a neural network. Our approach circumvents limiting factors like shape discretization and parameterization, as the proposed texture representation is independent of the shape representation of the 3D object. We show that Texture Fields are able to represent high frequency texture and naturally blend with modern deep learning techniques. Experimentally, we find that Texture Fields compare favorably to state-of-the-art methods for conditional texture reconstruction of 3D objects and enable learning of probabilistic generative models for texturing unseen 3D models. We believe that Texture Fields will become an important building block for the next generation of generative 3D models.

cs.CV↗

arXiv2018

Unpaired High-Resolution and Scalable Style Transfer Using Generative Adversarial Networks

Neural networks have proven their capabilities by outperforming many other approaches on regression or classification tasks on various kinds of data. Other astonishing results have been achieved using neural nets as data generators, especially in settings of generative adversarial networks (GANs). One special application is the field of image domain translations. Here, the goal is to take an image with a certain style (e.g. a photography) and transform it into another one (e.g. a painting). If such a task is performed for unpaired training examples, the corresponding GAN setting is complex, the neural networks are large, and this leads to a high peak memory consumption during, both, training and evaluation phase. This sets a limit to the highest processable image size. We address this issue by the idea of not processing the whole image at once, but to train and evaluate the domain translation on the level of overlapping image subsamples. This new approach not only enables us to translate high-resolution images that otherwise cannot be processed by the neural network at once, but also allows us to work with comparably small neural networks and with limited hardware resources. Additionally, the number of images required for the training process is significantly reduced. We present high-quality results on images with a total resolution of up to over 50 megapixels and emonstrate that our method helps to preserve local image details while it also keeps global consistency.

cs.CV↗

arXiv2018

Alternatives for Generating a Reduced Basis to Solve the Hyperspectral Diffuse Optical Tomography Model

The Reduced Basis Method (RBM) is a model reduction technique used to solve parametric PDEs that relies upon a basis set of solutions to the PDE at specific parameter values. To generate this reduced basis, the set of a small number of parameter values must be strategically chosen. We apply a Metropolis algorithm and a gradient algorithm to find the set of parameters and compare them to the standard greedy algorithm most commonly used in the RBM. We test our methods by using the RBM to solve a simplified version of the governing partial differential equation for hyperspectral diffuse optical tomography (hyDOT). The governing equation for hyDOT is an elliptic PDE parameterized by the wavelength of the laser source. For this one-dimensional problem, we find that both the Metropolis and gradient algorithms are potentially superior alternatives to the greedy algorithm in that they generate a reduced basis which produces solutions with a smaller relative error with respect to solutions found using the finite element method and in less time.

math.NA↗

arXiv2018

Ensemble Methods as a Defense to Adversarial Perturbations Against Deep Neural Networks

Deep learning has become the state of the art approach in many machine learning problems such as classification. It has recently been shown that deep learning is highly vulnerable to adversarial perturbations. Taking the camera systems of self-driving cars as an example, small adversarial perturbations can cause the system to make errors in important tasks, such as classifying traffic signs or detecting pedestrians. Hence, in order to use deep learning without safety concerns a proper defense strategy is required. We propose to use ensemble methods as a defense strategy against adversarial perturbations. We find that an attack leading one model to misclassify does not imply the same for other networks performing the same task. This makes ensemble methods an attractive defense strategy against adversarial attacks. We empirically show for the MNIST and the CIFAR-10 data sets that ensemble methods not only improve the accuracy of neural networks on test data but also increase their robustness against adversarial perturbations.

stat.ML↗