Search

Your search keyword '"Wagner, Moritz"' showing total 638 results
Start Over Author "Wagner, Moritz"
638 results on '"Wagner, Moritz"'

3. S-CFE: Simple Counterfactual Explanations

Author

Sadiku, Shpresim, Wagner, Moritz, Nagarajan, Sai Ganesh, and Pokutta, Sebastian

Subjects
Computer Science - Machine Learning, Mathematics - Optimization and Control
Abstract

We study the problem of finding optimal sparse, manifold-aligned counterfactual explanations for classifiers. Canonically, this can be formulated as an optimization problem with multiple non-convex components, including classifier loss functions and manifold alignment (or \emph{plausibility}) metrics. The added complexity of enforcing \emph{sparsity}, or shorter explanations, complicates the problem further. Existing methods often focus on specific models and plausibility measures, relying on convex $\ell_1$ regularizers to enforce sparsity. In this paper, we tackle the canonical formulation using the accelerated proximal gradient (APG) method, a simple yet efficient first-order procedure capable of handling smooth non-convex objectives and non-smooth $\ell_p$ (where $0 \leq p < 1$) regularizers. This enables our approach to seamlessly incorporate various classifiers and plausibility measures while producing sparser solutions. Our algorithm only requires differentiable data-manifold regularizers and supports box constraints for bounded feature ranges, ensuring the generated counterfactuals remain \emph{actionable}. Finally, experiments on real-world datasets demonstrate that our approach effectively produces sparse, manifold-aligned counterfactual explanations while maintaining proximity to the factual data and computational efficiency.

Published
2024

7. GSE: Group-wise Sparse and Explainable Adversarial Attacks

Author

Sadiku, Shpresim, Wagner, Moritz, and Pokutta, Sebastian

Subjects
Computer Science - Computer Vision and Pattern Recognition, Computer Science - Cryptography and Security, Computer Science - Machine Learning, Mathematics - Optimization and Control
Abstract

Sparse adversarial attacks fool deep neural networks (DNNs) through minimal pixel perturbations, often regularized by the $\ell_0$ norm. Recent efforts have replaced this norm with a structural sparsity regularizer, such as the nuclear group norm, to craft group-wise sparse adversarial attacks. The resulting perturbations are thus explainable and hold significant practical relevance, shedding light on an even greater vulnerability of DNNs. However, crafting such attacks poses an optimization challenge, as it involves computing norms for groups of pixels within a non-convex objective. We address this by presenting a two-phase algorithm that generates group-wise sparse attacks within semantically meaningful areas of an image. Initially, we optimize a quasinorm adversarial loss using the $1/2-$quasinorm proximal operator tailored for non-convex programming. Subsequently, the algorithm transitions to a projected Nesterov's accelerated gradient descent with $2-$norm regularization applied to perturbation magnitudes. Rigorous evaluations on CIFAR-10 and ImageNet datasets demonstrate a remarkable increase in group-wise sparsity, e.g., $50.9\%$ on CIFAR-10 and $38.4\%$ on ImageNet (average case, targeted attack). This performance improvement is accompanied by significantly faster computation times, improved explainability, and a $100\%$ attack success rate.

Published
2023

18. Deep Semi-Supervised Learning for Time Series Classification

Author

Goschenhofer, Jann, Hvingelby, Rasmus, Rügamer, David, Thomas, Janek, Wagner, Moritz, and Bischl, Bernd

Subjects
Computer Science - Machine Learning
Abstract

While Semi-supervised learning has gained much attention in computer vision on image data, yet limited research exists on its applicability in the time series domain. In this work, we investigate the transferability of state-of-the-art deep semi-supervised models from image to time series classification. We discuss the necessary model adaptations, in particular an appropriate model backbone architecture and the use of tailored data augmentation strategies. Based on these adaptations, we explore the potential of deep semi-supervised learning in the context of time series classification by evaluating our methods on large public time series classification problems with varying amounts of labelled samples. We perform extensive comparisons under a decidedly realistic and appropriate evaluation scheme with a unified reimplementation of all algorithms considered, which is yet lacking in the field. We find that these transferred semi-supervised models show significant performance gains over strong supervised, semi-supervised and self-supervised alternatives, especially for scenarios with very few labelled samples.

Published
2021
Full Text
View/download PDF

28. Changes of spino-pelvic characteristics post-THA are independent of surgical approach: a prospective study.

Author

Wagner, Moritz, Verhaegen, Jeroen, Vorimore, Camille, Innmann, Moritz, and Grammatopoulos, George

Subjects
*STATISTICAL significance, *HIP osteoarthritis, *MEDICAL sciences, *LONGITUDINAL method, *LORDOSIS
Abstract

Aims: Spinopelvic characteristics change after THA. Whether this change varies between approaches, is of interest for pre-op cup orientation planning. The aims of this study were to (1) Characterize changes in standing PT amongst patients with hip osteoarthritis treated with THA; (2) Test whether certain patient-related factors may predict PT change; and (3) Assess the association between surgical approach and PT change. Methods: This was a prospective, two-center, radiographic outcome study including 424 hips, consisting of anterior approach for 171 (40.3%) hips, lateral approach for 181 hips (42.7%) and posterior approach for 72 hips (17.0%). Spinopelvic characteristics were determined from lateral radiographs (before and one year after THA). Parameters of interest included: Pelvic tilt (PT), lumbar lordosis, sacral slope, pelvic incidence. PT change of more than 7 ° was considered clinically relevant. Results: Pelvic tilt increased by 2 ° from 15.1 ° (± 8.9) to 17.1 ° (± 9.7) after THA (p < 0.001). 19 hips (4.5%) experienced a relevant PT decrease, 337 (79.5%) had no clinically significant change in pelvic tilt, and 68 (16.0%) showed a moderate increase. Age, female sex and preoperative spinopelvic parameters including PT, SS and PI were predictive of PT change more than 7 °. PT increased most with lateral approach (2.9 ± 6.2) and least with anterior approach (1.1 ± 6.2, p = 0.024). Conclusion: Preoperative PT is the best predictor for PT change. PT is generally normalizing after THA and patients with low PT due to hip flexion contractures tend to increase PT after THA, few patients with high PT will decrease after THA. Anterior approach with capsulectomy was associated with the least change in PT post-THA. However, the approach-specific changes, although statistically significant, were too small to be considered during clinical practice, therefore no approach-specific prediction of PT change needs to be considered during preoperative planning for primary THA. [ABSTRACT FROM AUTHOR]

Published
2025
Full Text
View/download PDF

29. Deep-Learning-Based Reconstruction of Single-Breath-Hold 3 mm HASTE Improves Abdominal Image Quality and Reduces Acquisition Time: A Quantitative Analysis.

Author

Kubicka, Felix, Tan, Qinxuan, Meyer, Tom, Nickel, Dominik, Weiland, Elisabeth, Wagner, Moritz, and Marticorena Garcia, Stephan Rodrigo

Subjects
MAGNETIC resonance imaging, ADRENAL glands, PANCREATIC duct, DEEP learning, LIKERT scale
Abstract

Purpose: Breath-hold T2-weighted half-Fourier acquisition single-shot turbo spin echo (HASTE) magnetic resonance imaging (MRI) of the upper abdomen with a slice thickness below 5 mm suffers from high image noise and blurring. The purpose of this prospective study was to improve image quality and accelerate imaging acquisition by using single-breath-hold T2-weighted HASTE with deep learning (DL) reconstruction (DL-HASTE) with a 3 mm slice thickness. Method: MRI of the upper abdomen with DL-HASTE was performed in 35 participants (5 healthy volunteers and 30 patients) at 3 Tesla. In a subgroup of five healthy participants, signal-to-noise ratio (SNR) analysis was used after DL reconstruction to identify the smallest possible layer thickness (1, 2, 3, 4, 5 mm). DL-HASTE was acquired with a 3 mm slice thickness (DL-HASTE-3 mm) in 30 patients and compared with 5 mm DL-HASTE (DL-HASTE-5 mm) and with standard HASTE (standard-HASTE-5 mm). Image quality and motion artifacts were assessed quantitatively using Laplacian variance and semi-quantitatively by two radiologists using five-point Likert scales. Results: In the five healthy participants, DL-HASTE-3 mm was identified as the optimal slice (SNR 23.227 ± 3.901). Both DL-HASTE-3 mm and DL-HASTE-5 mm were assigned significantly higher overall image quality scores than standard-HASTE-5 mm (Laplacian variance, both p < 0.001; Likert scale, p < 0.001). Compared with DL-HASTE-5 mm (1.10 × 10−5 ± 6.93 × 10−6), DL-HASTE-3 mm (1.56 × 10−5 ± 8.69 × 10−6) provided a significantly higher SNR Laplacian variance (p < 0.001) and sharpness sub-scores for the intestinal tract, adrenal glands, and small anatomic structures (bile ducts, pancreatic ducts, and vessels; p < 0.05). Lesion detectability was rated excellent for both DL-HASTE-3 mm and DL-HASTE-5 mm (both: 5 [IQR4–5]) and was assigned higher scores than standard-HASTE-5 mm (4 [IQR4–5]; p < 0.001). DL-HASTE reduced the acquisition time by 63–69% compared with standard-HASTE-5 mm (p < 0.001). Conclusions: DL-HASTE is a robust abdominal MRI technique that improves image quality while at the same time reducing acquisition time compared with the routine clinical HASTE sequence. Using ultra-thin DL-HASTE-3 mm results in an even greater improvement with a similar SNR. [ABSTRACT FROM AUTHOR]

Published
2025
Full Text
View/download PDF

38. Distanz/Gegenwelt

Author

Wagner, Moritz, Weber, Ulrich, editor, Mauz, Andreas, editor, Stingelin, Martin, editor, Schmid, Kathrin, With Contrib. by, Morgenthaler, Simon, With Contrib. by, Thimm, Benjamin, With Contrib. by, and Schimchen, Philip, With Contrib. by

Published
2020
Full Text
View/download PDF

39. Frank der Fünfte

Author

Wagner, Moritz, Weber, Ulrich, editor, Mauz, Andreas, editor, Stingelin, Martin, editor, Schmid, Kathrin, With Contrib. by, Morgenthaler, Simon, With Contrib. by, Thimm, Benjamin, With Contrib. by, and Schimchen, Philip, With Contrib. by

Published
2020
Full Text
View/download PDF

Catalog

Books, media, physical & digital resources
See catalog results