Joachim Denzler

Aishwarya Venkataramanan, Sai Karthikeya Vemuri, Joachim Denzler

Physics models are inherently imperfect due to misspecified or missing mechanisms, resulting in systematic discrepancies between model predictions and real-world observations. The Kennedy-O'Hagan (KOH) framework addresses this issue through explicit discrepancy modeling. However, its non-amortized, per-instance formulation limits scalability across families of related systems. We introduce Amortized Physics-Informed Calibration (APIC), a population-level extension of KOH that leverages Neural Processes to perform scalable Bayesian inference across realizations. Our framework employs a two-branch latent architecture to disentangle instance-specific physical parameters from shared, state-dependent structural discrepancies. By integrating differentiable physics into an amortized inference backbone, APIC enables rapid calibration of unseen realizations from sparse observations while quantifying uncertainty. Experiments on the damped spring oscillator, the Lotka-Volterra system, and the advection-diffusion PDE with misspecified physics demonstrate improved parameter recovery and consistent identification of the systemic discrepancy structure compared to other calibration approaches.

Dimitri Korsch, Paul Bodesheim, Joachim Denzler

Biodiversity monitoring is crucial for tracking and counteracting adverse trends in population fluctuations. However, automatic recognition systems are rarely applied so far, and experts evaluate the generated data masses manually. Especially the support of deep learning methods for visual monitoring is not yet established in biodiversity research, compared to other areas like advertising or entertainment. In this paper, we present a deep learning pipeline for analyzing images captured by a moth scanner, an automated visual monitoring system of moth species developed within the AMMOD project. We first localize individuals with a moth detector and afterward determine the species of detected insects with a classifier. Our detector achieves up to 99.01% mean average precision and our classifier distinguishes 200 moth species with an accuracy of 93.13% on image cutouts depicting single insects. Combining both in our pipeline improves the accuracy for species identification in images of the moth scanner from 79.62% to 88.05%.

Dimitri Korsch, Paul Bodesheim, Gunnar Brehm et al.

Automatic camera-assisted monitoring of insects for abundance estimations is crucial to understand and counteract ongoing insect decline. In this paper, we present two datasets of nocturnal insects, especially moths as a subset of Lepidoptera, photographed in Central Europe. One of the datasets, the EU-Moths dataset, was captured manually by citizen scientists and contains species annotations for 200 different species and bounding box annotations for those. We used this dataset to develop and evaluate a two-stage pipeline for insect detection and moth species classification in previous work. We further introduce a prototype for an automated visual monitoring system. This prototype produced the second dataset consisting of more than 27,000 images captured on 95 nights. For evaluation and bootstrapping purposes, we annotated a subset of the images with bounding boxes enframing nocturnal insects. Finally, we present first detection and classification baselines for these datasets and encourage other scientists to use this publicly available data.

Dimitri Korsch, Maha Shadaydeh, Joachim Denzler

Fine-grained classification is a particular case of a classification problem, aiming to classify objects that share the visual appearance and can only be distinguished by subtle differences. Fine-grained classification models are often deployed to determine animal species or individuals in automated animal monitoring systems. Precise visual explanations of the model's decision are crucial to analyze systematic errors. Attention- or gradient-based methods are commonly used to identify regions in the image that contribute the most to the classification decision. These methods deliver either too coarse or too noisy explanations, unsuitable for identifying subtle visual differences reliably. However, perturbation-based methods can precisely identify pixels causally responsible for the classification result. Fill-in of the dropout (FIDO) algorithm is one of those methods. It utilizes the concrete dropout (CD) to sample a set of attribution masks and updates the sampling parameters based on the output of the classification model. A known problem of the algorithm is a high variance in the gradient estimates, which the authors have mitigated until now by mini-batch updates of the sampling parameters. This paper presents a solution to circumvent these computational instabilities by simplifying the CD sampling and reducing reliance on large mini-batch sizes. First, it allows estimating the parameters with smaller mini-batch sizes without losing the quality of the estimates but with a reduced computational effort. Furthermore, our solution produces finer and more coherent attribution masks. Finally, we use the resulting attribution masks to improve the classification performance of a trained model without additional fine-tuning of the model.

Matthias Körschens, Solveig Franziska Bucher, Christine Römermann et al.

The plant community composition is an essential indicator of environmental changes and is, for this reason, usually analyzed in ecological field studies in terms of the so-called plant cover. The manual acquisition of this kind of data is time-consuming, laborious, and prone to human error. Automated camera systems can collect high-resolution images of the surveyed vegetation plots at a high frequency. In combination with subsequent algorithmic analysis, it is possible to objectively extract information on plant community composition quickly and with little human effort. An automated camera system can easily collect the large amounts of image data necessary to train a Deep Learning system for automatic analysis. However, due to the amount of work required to annotate vegetation images with plant cover data, only few labeled samples are available. As automated camera systems can collect many pictures without labels, we introduce an approach to interpolate the sparse labels in the collected vegetation plot time series down to the intermediate dense and unlabeled images to artificially increase our training dataset to seven times its original size. Moreover, we introduce a new method we call Monte-Carlo Cropping. This approach trains on a collection of cropped parts of the training images to deal with high-resolution images efficiently, implicitly augment the training images, and speed up training. We evaluate both approaches on a plant cover dataset containing images of herbaceous plant communities and find that our methods lead to improvements in the species, community, and segmentation metrics investigated.

Tim Büchner, Sven Sickert, Gerd Fabian Volk et al.

The human face is one of the most crucial parts in interhuman communication. Even when parts of the face are hidden or obstructed the underlying facial movements can be understood. Machine learning approaches often fail in that regard due to the complexity of the facial structures. To alleviate this problem a common approach is to fine-tune a model for such a specific application. However, this is computational intensive and might have to be repeated for each desired analysis task. In this paper, we propose to reconstruct obstructed facial parts to avoid the task of repeated fine-tuning. As a result, existing facial analysis methods can be used without further changes with respect to the data. In our approach, the restoration of facial features is interpreted as a style transfer task between different recording setups. By using the CycleGAN architecture the requirement of matched pairs, which is often hard to fullfill, can be eliminated. To proof the viability of our approach, we compare our reconstructions with real unobstructed recordings. We created a novel data set in which 36 test subjects were recorded both with and without 62 surface electromyography sensors attached to their faces. In our evaluation, we feature typical facial analysis tasks, like the computation of Facial Action Units and the detection of emotions. To further assess the quality of the restoration, we also compare perceptional distances. We can show, that scores similar to the videos without obstructing sensors can be achieved.

Ferdinand Rewicki, Joachim Denzler, Julia Niebling

Detecting anomalies in time series data is important in a variety of fields, including system monitoring, healthcare, and cybersecurity. While the abundance of available methods makes it difficult to choose the most appropriate method for a given application, each method has its strengths in detecting certain types of anomalies. In this study, we compare six unsupervised anomaly detection methods of varying complexity to determine whether more complex methods generally perform better and if certain methods are better suited to certain types of anomalies. We evaluated the methods using the UCR anomaly archive, a recent benchmark dataset for anomaly detection. We analyzed the results on a dataset and anomaly type level after adjusting the necessary hyperparameters for each method. Additionally, we assessed the ability of each method to incorporate prior knowledge about anomalies and examined the differences between point-wise and sequence-wise features. Our experiments show that classical machine learning methods generally outperform deep learning methods across a range of anomaly types.

Sai Karthikeya Vemuri, Tim Büchner, Julia Niebling et al.

Physics-Informed Neural Networks (PINNs) have shown continuous and increasing promise in approximating partial differential equations (PDEs), although they remain constrained by the curse of dimensionality. In this paper, we propose a generalized PINN version of the classical variable separable method. To do this, we first show that, using the universal approximation theorem, a multivariate function can be approximated by the outer product of neural networks, whose inputs are separated variables. We leverage tensor decomposition forms to separate the variables in a PINN setting. By employing Canonic Polyadic (CP), Tensor-Train (TT), and Tucker decomposition forms within the PINN framework, we create robust architectures for learning multivariate functions from separate neural networks connected by outer products. Our methodology significantly enhances the performance of PINNs, as evidenced by improved results on complex high-dimensional PDEs, including the 3d Helmholtz and 5d Poisson equations, among others. This research underscores the potential of tensor decomposition-based variably separated PINNs to surpass the state-of-the-art, offering a compelling solution to the dimensionality challenge in PDE approximation.

Wasim Ahmad, Maha Shadaydeh, Joachim Denzler

Cause-effect analysis is crucial to understand the underlying mechanism of a system. We propose to exploit model invariance through interventions on the predictors to infer causality in nonlinear multivariate systems of time series. We model nonlinear interactions in time series using DeepAR and then expose the model to different environments using Knockoffs-based interventions to test model invariance. Knockoff samples are pairwise exchangeable, in-distribution and statistically null variables generated without knowing the response. We test model invariance where we show that the distribution of the response residual does not change significantly upon interventions on non-causal predictors. We evaluate our method on real and synthetically generated time series. Overall our method outperforms other widely used causality methods, i.e, VAR Granger causality, VARLiNGAM and PCMCI+.

Violeta Teodora Trifunov, Maha Shadaydeh, Joachim Denzler

Latent variables often mask cause-effect relationships in observational data which provokes spurious links that may be misinterpreted as causal. This problem sparks great interest in the fields such as climate science and economics. We propose to estimate confounded causal links of time series using Sequential Causal Effect Variational Autoencoder (SCEVAE) while applying Knockoff interventions. Knockoff variables have the same distribution as the originals and preserve the correlation to other variables. This allows for counterfactuals that are more faithful to the observational distribution. We show the advantage of Knockoff interventions by applying SCEVAE to synthetic datasets with both linear and nonlinear causal links. Moreover, we apply SCEVAE with Knockoffs to real aerosol-cloud-climate observational time series data. We compare our results on synthetic data to those of a time series deconfounding method both with and without estimated confounders. We show that our method outperforms this benchmark by comparing both methods to the ground truth. For the real data analysis, we rely on expert knowledge of causal links and demonstrate how using suitable proxy variables improves the causal link estimation in the presence of hidden confounders.

Tim Büchner, Niklas Penzel, Orlando Guntinas-Lichius et al.

Understanding expressions is vital for deciphering human behavior, and nowadays, end-to-end trained black box models achieve high performance. Due to the black-box nature of these models, it is unclear how they behave when applied out-of-distribution. Specifically, these models show decreased performance for unilateral facial palsy patients. We hypothesize that one crucial factor guiding the internal decision rules is facial symmetry. In this work, we use insights from causal reasoning to investigate the hypothesis. After deriving a structural causal model, we develop a synthetic interventional framework. This approach allows us to analyze how facial symmetry impacts a network's output behavior while keeping other factors fixed. All 17 investigated expression classifiers significantly lower their output activations for reduced symmetry. This result is congruent with observed behavior on real-world data from healthy subjects and facial palsy patients. As such, our investigation serves as a case study for identifying causal factors that influence the behavior of black-box models.

Dong Han, Yufan Jiang, Yong Li et al.

In this work, we leverage the pure skin color patch from the face image as the additional information to train an auxiliary skin color feature extractor and face recognition model in parallel to improve performance of state-of-the-art (SOTA) privacy-preserving face recognition (PPFR) systems. Our solution is robust against black-box attacking and well-established generative adversarial network (GAN) based image restoration. We analyze the potential risk in previous work, where the proposed cosine similarity computation might directly leak the protected precomputed embedding stored on the server side. We propose a Function Secret Sharing (FSS) based face embedding comparison protocol without any intermediate result leakage. In addition, we show in experiments that the proposed protocol is more efficient compared to the Secret Sharing (SS) based protocol.

Felix Schneider, Maria Gogolev, Sven Sickert et al.

Tokenization and sub-tokenization based models like word2vec, BERT and the GPTs are the state-of-the-art in natural language processing. Typically, these approaches have limitations with respect to their input representation. They fail to fully capture orthographic similarities and morphological variations, especially in highly inflected and under-resource languages. To mitigate this problem, we propose to computes word vectors directly from character strings, integrating both semantic and syntactic information. We denote this transformer-based approach Rich Character Embeddings (RCE). Furthermore, we propose a hybrid model that combines transformer and convolutional mechanisms. Both vector representations can be used as a drop-in replacement for dictionary- and subtoken-based word embeddings in existing model architectures. It has the potential to improve performance for both large context-based language models like BERT and small models like word2vec for under-resourced and morphologically rich languages. We evaluate our approach on various tasks like the SWAG, declension prediction for inflected languages, metaphor and chiasmus detection for various languages. Our experiments show that it outperforms traditional token-based approaches on limited data using OddOneOut and TopK metrics.

Dong Han, Yong Li, Joachim Denzler

With the advancement of face recognition (FR) systems, privacy-preserving face recognition (PPFR) systems have gained popularity for their accurate recognition, enhanced facial privacy protection, and robustness to various attacks. However, there are limited studies to further verify privacy risks by reconstructing realistic high-resolution face images from embeddings of these systems, especially for PPFR. In this work, we propose the face embedding mapping (FEM), a general framework that explores Kolmogorov-Arnold Network (KAN) for conducting the embedding-to-face attack by leveraging pre-trained Identity-Preserving diffusion model against state-of-the-art (SOTA) FR and PPFR systems. Based on extensive experiments, we verify that reconstructed faces can be used for accessing other real-word FR systems. Besides, the proposed method shows the robustness in reconstructing faces from the partial and protected face embeddings. Moreover, FEM can be utilized as a tool for evaluating safety of FR and PPFR systems in terms of privacy leakage. All images used in this work are from public datasets.

Wasim Ahmad, Maha Shadaydeh, Joachim Denzler

Causal inference in a nonlinear system of multivariate timeseries is instrumental in disentangling the intricate web of relationships among variables, enabling us to make more accurate predictions and gain deeper insights into real-world complex systems. Causality methods typically identify the causal structure of a multivariate system by considering the cause-effect relationship of each pair of variables while ignoring the collective effect of a group of variables or interactions involving more than two-time series variables. In this work, we test model invariance by group-level interventions on the trained deep networks to infer causal direction in groups of variables, such as climate and ecosystem, brain networks, etc. Extensive testing with synthetic and real-world time series data shows a significant improvement of our method over other applied group causality methods and provides us insights into real-world time series. The code for our method can be found at:https://github.com/wasimahmadpk/gCause.

Jonas Jäger, Gereon Reus, Joachim Denzler et al.

State-of-the-art computer vision approaches rely on huge amounts of annotated data. The collection of such data is a time consuming process since it is mainly performed by humans. The literature shows that semi-automatic annotation approaches can significantly speed up the annotation process by the automatic generation of annotation proposals to support the annotator. In this paper we present a framework that allows for a quick and flexible design of semi-automatic annotation pipelines. We show that a good design of the process will speed up the collection of annotations. Our contribution is a new approach to image annotation that allows for the combination of different annotation tools and machine learning algorithms in one process. We further present potential applications of our approach. The source code of our framework called LOST (Label Objects and Save Time) is available at: https://github.com/l3p-cv/lost.

Oliver Mothes, Joachim Denzler

The essential task of verifying person identities at airports and national borders is very time consuming. To accelerate it, optical character recognition for identity documents (IDs) using dictionaries is not appropriate due to high variability of the text content in IDs, e.g., individual street names or surnames. Additionally, no properties of the used fonts in IDs are known. Therefore, we propose an iterative self-supervised bootstrapping approach using a smart strategy to mine real character data from IDs. In combination with synthetically generated character data, the real data is used to train efficient convolutional neural networks for character classification serving a practical runtime as well as a high accuracy. On a dataset with 74 character classes, we achieve an average class-wise accuracy of 99.4 %. In contrast, if we would apply a classifier trained only using synthetic data, the accuracy is reduced to 58.1 %. Finally, we show that our whole proposed pipeline outperforms an established open-source framework

Marcel Simon, Erik Rodner, Joachim Denzler

Convolutional neural networks (CNN) pre-trained on ImageNet are the backbone of most state-of-the-art approaches. In this paper, we present a new set of pre-trained models with popular state-of-the-art architectures for the Caffe framework. The first release includes Residual Networks (ResNets) with generation script as well as the batch-normalization-variants of AlexNet and VGG19. All models outperform previous models with the same architecture. The models and training code are available at http://www.inf-cv.uni-jena.de/Research/CNN+Models.html and https://github.com/cvjena/cnn-models

Marcel Simon, Erik Rodner, Joachim Denzler

Current fine-grained classification approaches often rely on a robust localization of object parts to extract localized feature representations suitable for discrimination. However, part localization is a challenging task due to the large variation of appearance and pose. In this paper, we show how pre-trained convolutional neural networks can be used for robust and efficient object part discovery and localization without the necessity to actually train the network on the current dataset. Our approach called "part detector discovery" (PDD) is based on analyzing the gradient maps of the network outputs and finding activation centers spatially related to annotated semantic parts or bounding boxes. This allows us not just to obtain excellent performance on the CUB200-2011 dataset, but in contrast to previous approaches also to perform detection and bird classification jointly without requiring a given bounding box annotation during testing and ground-truth parts during training. The code is available at http://www.inf-cv.uni-jena.de/part_discovery and https://github.com/cvjena/PartDetectorDisovery.

Gideon Stein, Niklas Penzel, Tristan Piater et al.

Causal Discovery (CD) is a powerful framework for scientific inquiry. Yet, its practical adoption is hindered by a reliance on strong, often unverifiable assumptions and a lack of robust performance assessment. To address these limitations and advance empirical CD evaluation, we present TCD-Arena, a modularized, highly customizable, and extendable testing kit to assess the robustness of time series CD algorithms against stepwise more severe assumption violations. For demonstration, we conduct an extensive empirical study comprising around 30 million individual CD attempts and reveal nuanced robustness profiles for 33 distinct assumption violations. Further, we investigate CD ensembles and find that they have the potential to improve general robustness, which has implications for real-world applications. With this, we strive to ultimately facilitate the development of CD methods that are reliable for a diverse range of synthetic and potentially real-world data conditions.

Gideon Stein, Maha Shadaydeh, Jan Blunk et al.

Causal discovery, or identifying causal relationships from observational data, is a notoriously challenging task, with numerous methods proposed to tackle it. Despite this, in-the-wild evaluation of these methods is still lacking, as works frequently rely on synthetic data evaluation and sparse real-world examples under critical theoretical assumptions. Real-world causal structures, however, are often complex, making it hard to decide on a proper causal discovery strategy. To bridge this gap, we introduce CausalRivers, the largest in-the-wild causal discovery benchmarking kit for time-series data to date. CausalRivers features an extensive dataset on river discharge that covers the eastern German territory (666 measurement stations) and the state of Bavaria (494 measurement stations). It spans the years 2019 to 2023 with a 15-minute temporal resolution. Further, we provide additional data from a flood around the Elbe River, as an event with a pronounced distributional shift. Leveraging multiple sources of information and time-series meta-data, we constructed two distinct causal ground truth graphs (Bavaria and eastern Germany). These graphs can be sampled to generate thousands of subgraphs to benchmark causal discovery across diverse and challenging settings. To demonstrate the utility of CausalRivers, we evaluate several causal discovery approaches through a set of experiments to identify areas for improvement. CausalRivers has the potential to facilitate robust evaluations and comparisons of causal discovery methods. Besides this primary purpose, we also expect that this dataset will be relevant for connected areas of research, such as time-series forecasting and anomaly detection. Based on this, we hope to push benchmark-driven method development that fosters advanced techniques for causal discovery, as is the case for many other areas of machine learning.

Dong Han, Yong Li, Joachim Denzler

With the advancement of face reconstruction (FR) systems, privacy-preserving face recognition (PPFR) has gained popularity for its secure face recognition, enhanced facial privacy protection, and robustness to various attacks. Besides, specific models and algorithms are proposed for face embedding protection by mapping embeddings to a secure space. However, there is a lack of studies on investigating and evaluating the possibility of extracting face images from embeddings of those systems, especially for PPFR. In this work, we introduce the first approach to exploit Kolmogorov-Arnold Network (KAN) for conducting embedding-to-face attacks against state-of-the-art (SOTA) FR and PPFR systems. Face embedding mapping (FEM) models are proposed to learn the distribution mapping relation between the embeddings from the initial domain and target domain. In comparison with Multi-Layer Perceptrons (MLP), we provide two variants, FEM-KAN and FEM-MLP, for efficient non-linear embedding-to-embedding mapping in order to reconstruct realistic face images from the corresponding face embedding. To verify our methods, we conduct extensive experiments with various PPFR and FR models. We also measure reconstructed face images with different metrics to evaluate the image quality. Through comprehensive experiments, we demonstrate the effectiveness of FEMs in accurate embedding mapping and face reconstruction.

Tim Büchner, Niklas Penzel, Orlando Guntinas-Lichius et al.

Facial expression-based human emotion recognition is a critical research area in psychology and medicine. State-of-the-art classification performance is only reached by end-to-end trained neural networks. Nevertheless, such black-box models lack transparency in their decision-making processes, prompting efforts to ascertain the rules that underlie classifiers' decisions. Analyzing single inputs alone fails to expose systematic learned biases. These biases can be characterized as facial properties summarizing abstract information like age or medical conditions. Therefore, understanding a model's prediction behavior requires an analysis rooted in causality along such selected properties. We demonstrate that up to 91.25% of classifier output behavior changes are statistically significant concerning basic properties. Among those are age, gender, and facial symmetry. Furthermore, the medical usage of surface electromyography significantly influences emotion prediction. We introduce a workflow to evaluate explicit properties and their impact. These insights might help medical professionals select and apply classifiers regarding their specialized data and properties.

Laines Schmalwasser, Jakob Gawlikowski, Joachim Denzler et al.

Concept Activation Vectors (CAVs) offer insights into neural network decision-making by linking human friendly concepts to the model's internal feature extraction process. However, when a new set of CAVs is discovered, they must still be translated into a human understandable description. For image-based neural networks, this is typically done by visualizing the most relevant images of a CAV, while the determination of the concept is left to humans. In this work, we introduce an approach to aid the interpretation of newly discovered concept sets by suggesting textual descriptions for each CAV. This is done by mapping the most relevant images representing a CAV into a text-image embedding where a joint description of these relevant images can be computed. We propose utilizing the most relevant receptive fields instead of full images encoded. We demonstrate the capabilities of this approach in multiple experiments with and without given CAV labels, showing that the proposed approach provides accurate descriptions for the CAVs and reduces the challenge of concept interpretation.

Laines Schmalwasser, Niklas Penzel, Joachim Denzler et al.

Concepts such as objects, patterns, and shapes are how humans understand the world. Building on this intuition, concept-based explainability methods aim to study representations learned by deep neural networks in relation to human-understandable concepts. Here, Concept Activation Vectors (CAVs) are an important tool and can identify whether a model learned a concept or not. However, the computational cost and time requirements of existing CAV computation pose a significant challenge, particularly in large-scale, high-dimensional architectures. To address this limitation, we introduce FastCAV, a novel approach that accelerates the extraction of CAVs by up to 63.6x (on average 46.4x). We provide a theoretical foundation for our approach and give concrete assumptions under which it is equivalent to established SVM-based methods. Our empirical results demonstrate that CAVs calculated with FastCAV maintain similar performance while being more efficient and stable. In downstream applications, i.e., concept-based explanation methods, we show that FastCAV can act as a replacement leading to equivalent insights. Hence, our approach enables previously infeasible investigations of deep models, which we demonstrate by tracking the evolution of concepts during model training.

Aishwarya Venkataramanan, Paul Bodesheim, Joachim Denzler

Vision-Language Models (VLMs) learn joint representations by mapping images and text into a shared latent space. However, recent research highlights that deterministic embeddings from standard VLMs often struggle to capture the uncertainties arising from the ambiguities in visual and textual descriptions and the multiple possible correspondences between images and texts. Existing approaches tackle this by learning probabilistic embeddings during VLM training, which demands large datasets and does not leverage the powerful representations already learned by large-scale VLMs like CLIP. In this paper, we propose GroVE, a post-hoc approach to obtaining probabilistic embeddings from frozen VLMs. GroVE builds on Gaussian Process Latent Variable Model (GPLVM) to learn a shared low-dimensional latent space where image and text inputs are mapped to a unified representation, optimized through single-modal embedding reconstruction and cross-modal alignment objectives. Once trained, the Gaussian Process model generates uncertainty-aware probabilistic embeddings. Evaluation shows that GroVE achieves state-of-the-art uncertainty calibration across multiple downstream tasks, including cross-modal retrieval, visual question answering, and active learning.

Tim Büchner, Christoph Anders, Orlando Guntinas-Lichius et al.

The relationship between muscle activity and resulting facial expressions is crucial for various fields, including psychology, medicine, and entertainment. The synchronous recording of facial mimicry and muscular activity via surface electromyography (sEMG) provides a unique window into these complex dynamics. Unfortunately, existing methods for facial analysis cannot handle electrode occlusion, rendering them ineffective. Even with occlusion-free reference images of the same person, variations in expression intensity and execution are unmatchable. Our electromyography-informed facial expression reconstruction (EIFER) approach is a novel method to restore faces under sEMG occlusion faithfully in an adversarial manner. We decouple facial geometry and visual appearance (e.g., skin texture, lighting, electrodes) by combining a 3D Morphable Model (3DMM) with neural unpaired image-to-image translation via reference recordings. Then, EIFER learns a bidirectional mapping between 3DMM expression parameters and muscle activity, establishing correspondence between the two domains. We validate the effectiveness of our approach through experiments on a dataset of synchronized sEMG recordings and facial mimicry, demonstrating faithful geometry and appearance reconstruction. Further, we synthesize expressions based on muscle activity and how observed expressions can predict dynamic muscle activity. Consequently, EIFER introduces a new paradigm for facial electromyography, which could be extended to other forms of multi-modal face recordings.

Gideon Stein, Maha Shadaydeh, Joachim Denzler

Causal discovery from time series data encompasses many existing solutions, including those based on deep learning techniques. However, these methods typically do not endorse one of the most prevalent paradigms in deep learning: End-to-end learning. To address this gap, we explore what we call Causal Pretraining. A methodology that aims to learn a direct mapping from multivariate time series to the underlying causal graphs in a supervised manner. Our empirical findings suggest that causal discovery in a supervised manner is possible, assuming that the training and test time series samples share most of their dynamics. More importantly, we found evidence that the performance of Causal Pretraining can increase with data and model size, even if the additional data do not share the same dynamics. Further, we provide examples where causal discovery for real-world data with causally pretrained neural networks is possible within limits. We argue that this hints at the possibility of a foundation model for causal discovery.

Tim Büchner, Oliver Mothes, Orlando Guntinas-Lichius et al.

Analyzing facial features and expressions is a complex task in computer vision. The human face is intricate, with significant shape, texture, and appearance variations. In medical contexts, facial structures and movements that differ from the norm are particularly important to study and require precise analysis to understand the underlying conditions. Given that solely the facial muscles, innervated by the facial nerve, are responsible for facial expressions, facial palsy can lead to severe impairments in facial movements. One affected area of interest is the subtle movements involved in blinking. It is an intricate spontaneous process that is not yet fully understood and needs high-resolution, time-specific analysis for detailed understanding. However, a significant challenge is that many computer vision techniques demand programming skills for automated extraction and analysis, making them less accessible to medical professionals who may not have these skills. The Jena Facial Palsy Toolbox (JeFaPaTo) has been developed to bridge this gap. It utilizes cutting-edge computer vision algorithms and offers a user-friendly interface for those without programming expertise. This toolbox makes advanced facial analysis more accessible to medical experts, simplifying integration into their workflow.

Sai Karthikeya Vemuri, Tim Büchner, Joachim Denzler

Implicit Neural Representation (INR) has emerged as a powerful tool for encoding discrete signals into continuous, differentiable functions using neural networks. However, these models often have an unfortunate reliance on monolithic architectures to represent high-dimensional data, leading to prohibitive computational costs as dimensionality grows. We propose F-INR, a framework that reformulates INR learning through functional tensor decomposition, breaking down high-dimensional tasks into lightweight, axis-specific sub-networks. Each sub-network learns a low-dimensional data component (e.g., spatial or temporal). Then, we combine these components via tensor operations, reducing forward pass complexity while improving accuracy through specialized learning. F-INR is modular and, therefore, architecture-agnostic, compatible with MLPs, SIREN, WIRE, or other state-of-the-art INR architecture. It is also decomposition-agnostic, supporting CP, TT, and Tucker modes with user-defined rank for speed-accuracy control. In our experiments, F-INR trains $100\times$ faster than existing approaches on video tasks while achieving higher fidelity (+3.4 dB PSNR). Similar gains hold for image compression, physics simulations, and 3D geometry reconstruction. Through this, F-INR offers a new scalable, flexible solution for high-dimensional signal modeling.

Tristan Piater, Niklas Penzel, Gideon Stein et al.

A substantial body of research has focused on developing systems that assist medical professionals during labor-intensive early screening processes, many based on convolutional deep-learning architectures. Recently, multiple studies explored the application of so-called self-attention mechanisms in the vision domain. These studies often report empirical improvements over fully convolutional approaches on various datasets and tasks. To evaluate this trend for medical imaging, we extend two widely adopted convolutional architectures with different self-attention variants on two different medical datasets. With this, we aim to specifically evaluate the possible advantages of additional self-attention. We compare our models with similarly sized convolutional and attention-based baselines and evaluate performance gains statistically. Additionally, we investigate how including such layers changes the features learned by these models during the training. Following a hyperparameter search, and contrary to our expectations, we observe no significant improvement in balanced accuracy over fully convolutional models. We also find that important features, such as dermoscopic structures in skin lesion images, are still not learned by employing self-attention. Finally, analyzing local explanations, we confirm biased feature usage. We conclude that merely incorporating attention is insufficient to surpass the performance of existing fully convolutional methods.

Aishwarya Venkataramanan, Joachim Denzler

We propose the Distance-informed Neural Process (DNP), a novel variant of Neural Processes that improves uncertainty estimation by combining global and distance-aware local latent structures. Standard Neural Processes (NPs) often rely on a global latent variable and struggle with uncertainty calibration and capturing local data dependencies. DNP addresses these limitations by introducing a global latent variable to model task-level variations and a local latent variable to capture input similarity within a distance-preserving latent space. This is achieved through bi-Lipschitz regularization, which bounds distortions in input relationships and encourages the preservation of relative distances in the latent space. This modeling approach allows DNP to produce better-calibrated uncertainty estimates and more effectively distinguish in- from out-of-distribution data. Empirical results demonstrate that DNP achieves strong predictive performance and improved uncertainty calibration across regression and classification tasks.

Ihab Asaad, Maha Shadaydeh, Joachim Denzler

Machine learning classification models trained with empirical risk minimization (ERM) often inadvertently rely on spurious correlations. When absent in the test data, these unintended associations between non-target attributes and target labels lead to poor generalization. This paper addresses this problem from a model optimization perspective and proposes a novel method, Gradient Extrapolation for Debiased Representation Learning (GERNE), designed to learn debiased representations in both known and unknown attribute training cases. GERNE uses two distinct batches with different amounts of spurious correlations and defines the target gradient as a linear extrapolation of the gradients computed from each batch's loss. Our analysis shows that when the extrapolated gradient points toward the batch gradient with fewer spurious correlations, it effectively guides training toward learning a debiased model. GERNE serves as a general framework for debiasing, encompassing ERM and Resampling methods as special cases. We derive the theoretical upper and lower bounds of the extrapolation factor employed by GERNE. By tuning this factor, GERNE can adapt to maximize either Group-Balanced Accuracy (GBA) or Worst-Group Accuracy (WGA). We validate GERNE on five vision and one NLP benchmarks, demonstrating competitive and often superior performance compared to state-of-the-art baselines. The project page is available at: https://gerne-debias.github.io/.

Ihab Asaad, Maha Shadaydeh, Joachim Denzler

Patch-wise multi-label classification provides an efficient alternative to full pixel-wise segmentation on high-resolution images, particularly when the objective is to determine the presence or absence of target objects within a patch rather than their precise spatial extent. This formulation substantially reduces annotation cost, simplifies training, and allows flexible patch sizing aligned with the desired level of decision granularity. In this work, we focus on a special case, patch-wise binary classification, applied to the detection of a single class of interest (tumor) on high-resolution multimodal nonlinear microscopy images. We show that, although this simplified formulation enables efficient model development, it can introduce spurious correlations between patch composition and labels: tumor patches tend to contain larger tissue regions, whereas non-tumor patches often consist mostly of background with small tissue areas. We further quantify the bias in model predictions caused by this spurious correlation, and propose to use a debiasing strategy to mitigate its effect. Specifically, we apply GERNE, a debiasing method that can be adapted to maximize worst-group accuracy (WGA). Our results show an improvement in WGA by approximately 7% compared to ERM for two different thresholds used to binarize the spurious feature. This enhancement boosts model performance on critical minority cases, such as tumor patches with small tissues and non-tumor patches with large tissues, and underscores the importance of spurious correlation-aware learning in patch-wise classification problems.

Aishwarya Venkataramanan, Sai Karthikeya Vemuri, Adithya Ashok Chalain Valapil et al.

Coherent anti-Stokes Raman scattering (CARS) spectroscopy is a powerful and rapid technique widely used in medicine, material science, and chemical analyses. However, its effectiveness is hindered by the presence of a non-resonant background that interferes with and distorts the true Raman signal. Deep learning methods have been employed to reconstruct the true Raman spectrum from measured CARS data using labeled datasets. A more recent development integrates the domain knowledge of Kramers-Kronig relationships and smoothness constraints in the form of physics-informed loss functions. However, these deterministic models lack the ability to quantify uncertainty, an essential feature for reliable deployment in high-stakes scientific and biomedical applications. In this work, we evaluate and compare various uncertainty quantification (UQ) techniques within the context of CARS-to-Raman signal reconstruction. Furthermore, we demonstrate that incorporating physics-informed constraints into these models improves their calibration, offering a promising path toward more trustworthy CARS data analysis.

Wasim Ahmad, Joachim Denzler, Maha Shadaydeh

Causal discovery uncovers complex relationships between variables, enhancing predictions, decision-making, and insights into real-world systems, especially in nonlinear multivariate time series. However, most existing methods primarily focus on pairwise cause-effect relationships, overlooking interactions among groups of variables, i.e., subsystems and their collective causal influence. In this study, we introduce gCDMI, a novel multi-group causal discovery method that leverages group-level interventions on trained deep neural networks and employs model invariance testing to infer causal relationships. Our approach involves three key steps. First, we use deep learning to jointly model the structural relationships among groups of all time series. Second, we apply group-wise interventions to the trained model. Finally, we conduct model invariance testing to determine the presence of causal links among variable groups. We evaluate our method on simulated datasets, demonstrating its superior performance in identifying group-level causal relationships compared to existing methods. Additionally, we validate our approach on real-world datasets, including brain networks and climate ecosystems. Our results highlight that applying group-level interventions to deep learning models, combined with invariance testing, can effectively reveal complex causal structures, offering valuable insights for domains such as neuroscience and climate science.

Phillip Rothenbeck, Sai Karthikeya Vemuri, Niklas Penzel et al.

The COVID-19 pandemic has highlighted the need for quantitative modeling and analysis to understand real-world disease dynamics. In particular, post hoc analyses using compartmental models offer valuable insights into the effectiveness of public health interventions, such as vaccination strategies and containment policies. However, such compartmental models like SIR (Susceptible-Infectious-Recovered) often face limitations in directly incorporating noisy observational data. In this work, we employ Physics-Informed Neural Networks (PINNs) to solve the inverse problem of the SIR model using infection data from the Robert Koch Institute (RKI). Our main contribution is a fine-grained, spatio-temporal analysis of COVID-19 dynamics across all German federal states over a three-year period. We estimate state-specific transmission and recovery parameters and time-varying reproduction number (R_t) to track the pandemic progression. The results highlight strong variations in transmission behavior across regions, revealing correlations with vaccination uptake and temporal patterns associated with major pandemic phases. Our findings demonstrate the utility of PINNs in localized, long-term epidemiological modeling.

Sai Karthikeya Vemuri, Adithya Ashok Chalain Valapil, Tim Büchner et al.

Transferring the recent advancements in deep learning into scientific disciplines is hindered by the lack of the required large-scale datasets for training. We argue that in these knowledge-rich domains, the established body of scientific theory provides reliable inductive biases in the form of governing physical laws. We address the ill-posed inverse problem of recovering Raman spectra from noisy Coherent Anti-Stokes Raman Scattering (CARS) measurements, as the true Raman signal here is suppressed by a dominating non-resonant background. We propose RamPINN, a model that learns to recover Raman spectra from given CARS spectra. Our core methodological contribution is a physics-informed neural network that utilizes a dual-decoder architecture to disentangle resonant and non-resonant signals. This is done by enforcing the Kramers-Kronig causality relations via a differentiable Hilbert transform loss on the resonant and a smoothness prior on the non-resonant part of the signal. Trained entirely on synthetic data, RamPINN demonstrates strong zero-shot generalization to real-world experimental data, explicitly closing this gap and significantly outperforming existing baselines. Furthermore, we show that training with these physics-based losses alone, without access to any ground-truth Raman spectra, still yields competitive results. This work highlights a broader concept: formal scientific rules can act as a potent inductive bias, enabling robust, self-supervised learning in data-limited scientific domains.

Johann Schmidt, Sebastian Stober, Joachim Denzler et al.

Fine-grained visual classification (FGVC) requires distinguishing between visually similar categories through subtle, localized features - a task that remains challenging due to high intra-class variability and limited inter-class differences. Existing part-based methods often rely on complex localization networks that learn mappings from pixel to sample space, requiring a deep understanding of image content while limiting feature utility for downstream tasks. In addition, sampled points frequently suffer from high spatial redundancy, making it difficult to quantify the optimal number of required parts. Inspired by human saccadic vision, we propose a two-stage process that first extracts peripheral features (coarse view) and generates a sample map, from which fixation patches are sampled and encoded in parallel using a weight-shared encoder. We employ contextualized selective attention to weigh the impact of each fixation patch before fusing peripheral and focus representations. To prevent spatial collapse - a common issue in part-based methods - we utilize non-maximum suppression during fixation sampling to eliminate redundancy. Comprehensive evaluation on standard FGVC benchmarks (CUB-200-2011, NABirds, Food-101 and Stanford-Dogs) and challenging insect datasets (EU-Moths, Ecuador-Moths and AMI-Moths) demonstrates that our method achieves comparable performance to state-of-the-art approaches while consistently outperforming our baseline encoder.

Niklas Penzel, Joachim Denzler

Deep learning models achieve high predictive performance but lack intrinsic interpretability, hindering our understanding of the learned prediction behavior. Existing local explainability methods focus on associations, neglecting the causal drivers of model predictions. Other approaches adopt a causal perspective but primarily provide global, model-level explanations. However, for specific inputs, it's unclear whether globally identified factors apply locally. To address this limitation, we introduce a novel framework for local interventional explanations by leveraging recent advances in image-to-image editing models. Our approach performs gradual interventions on semantic properties to quantify the corresponding impact on a model's predictions using a novel score, the expected property gradient magnitude. We demonstrate the effectiveness of our approach through an extensive empirical evaluation on a wide range of architectures and tasks. First, we validate it in a synthetic scenario and demonstrate its ability to locally identify biases. Afterward, we apply our approach to investigate medical skin lesion classifiers, analyze network training dynamics, and study a pre-trained CLIP model with real-life interventional data. Our results highlight the potential of interventional explanations on the property level to reveal new insights into the behavior of deep models.

Ferdinand Rewicki, Joachim Denzler, Julia Niebling

Detecting and classifying abnormal system states is critical for condition monitoring, but supervised methods often fall short due to the rarity of anomalies and the lack of labeled data. Therefore, clustering is often used to group similar abnormal behavior. However, evaluating cluster quality without ground truth is challenging, as existing measures such as the Silhouette Score (SSC) only evaluate the cohesion and separation of clusters and ignore possible prior knowledge about the data. To address this challenge, we introduce the Synchronized Anomaly Agreement Index (SAAI), which exploits the synchronicity of anomalies across multivariate time series to assess cluster quality. We demonstrate the effectiveness of SAAI by showing that maximizing SAAI improves accuracy on the task of finding the true number of anomaly classes K in correlated time series by 0.23 compared to SSC and by 0.32 compared to X-Means. We also show that clusters obtained by maximizing SAAI are easier to interpret compared to SSC.

Ferdinand Rewicki, Jakob Gawlikowski, Julia Niebling et al.

The detection of abnormal or critical system states is essential in condition monitoring. While much attention is given to promptly identifying anomalies, a retrospective analysis of these anomalies can significantly enhance our comprehension of the underlying causes of observed undesired behavior. This aspect becomes particularly critical when the monitored system is deployed in a vital environment. In this study, we delve into anomalies within the domain of Bio-Regenerative Life Support Systems (BLSS) for space exploration and analyze anomalies found in telemetry data stemming from the EDEN ISS space greenhouse in Antarctica. We employ time series clustering on anomaly detection results to categorize various types of anomalies in both uni- and multivariate settings. We then assess the effectiveness of these methods in identifying systematic anomalous behavior. Additionally, we illustrate that the anomaly detection methods MDI and DAMP produce complementary results, as previously indicated by research.

Niklas Penzel, Gideon Stein, Joachim Denzler

Deep models trained on large amounts of data often incorporate implicit biases present during training time. If later such a bias is discovered during inference or deployment, it is often necessary to acquire new data and retrain the model. This behavior is especially problematic in critical areas such as autonomous driving or medical decision-making. In these scenarios, new data is often expensive and hard to come by. In this work, we present a method based on change penalization that takes a pre-trained model and adapts the weights to mitigate a previously detected bias. We achieve this by tuning a zero-initialized copy of a frozen pre-trained network. Our method needs very few, in extreme cases only a single, examples that contradict the bias to increase performance. Additionally, we propose an early stopping criterion to modify baselines and reduce overfitting. We evaluate our approach on a well-known bias in skin lesion classification and three other datasets from the domain shift literature. We find that our approach works especially well with very few images. Simple fine-tuning combined with our early stopping also leads to performance benefits for a larger number of tuning samples.

Dong Han, Yong Li, Joachim Denzler

Face recognition technology has been deployed in various real-life applications. The most sophisticated deep learning-based face recognition systems rely on training millions of face images through complex deep neural networks to achieve high accuracy. It is quite common for clients to upload face images to the service provider in order to access the model inference. However, the face image is a type of sensitive biometric attribute tied to the identity information of each user. Directly exposing the raw face image to the service provider poses a threat to the user's privacy. Current privacy-preserving approaches to face recognition focus on either concealing visual information on model input or protecting model output face embedding. The noticeable drop in recognition accuracy is a pitfall for most methods. This paper proposes a hybrid frequency-color fusion approach to reduce the input dimensionality of face recognition in the frequency domain. Moreover, sparse color information is also introduced to alleviate significant accuracy degradation after adding differential privacy noise. Besides, an identity-specific embedding mapping scheme is applied to protect original face embedding by enlarging the distance among identities. Lastly, secure multiparty computation is implemented for safely computing the embedding distance during model inference. The proposed method performs well on multiple widely used verification datasets. Moreover, it has around 2.6% to 4.2% higher accuracy than the state-of-the-art in the 1:N verification scenario.

Bernd Gruner, Matthias Körschens, Björn Barz et al.

Deep-learning methods offer unsurpassed recognition performance in a wide range of domains, including fine-grained recognition tasks. However, in most problem areas there are insufficient annotated training samples. Therefore, the topic of transfer learning respectively domain adaptation is particularly important. In this work, we investigate to what extent unsupervised domain adaptation can be used for fine-grained recognition in a biodiversity context to learn a real-world classifier based on idealized training data, e.g. preserved butterflies and plants. Moreover, we investigate the influence of different normalization layers, such as Group Normalization in combination with Weight Standardization, on the classifier. We discovered that domain adaptation works very well for fine-grained recognition and that the normalization methods have a great influence on the results. Using domain adaptation and Transferable Normalization, the accuracy of the classifier could be increased by up to 12.35 % compared to the baseline. Furthermore, the domain adaptation system is combined with an active learning component to improve the results. We compare different active learning strategies with each other. Surprisingly, we found that more sophisticated strategies provide better results than the random selection baseline for only one of the two datasets. In this case, the distance and diversity strategy performed best. Finally, we present a problem analysis of the datasets.

Björn Barz, Lorenzo Brigato, Luca Iocchi et al.

Learning from limited amounts of data is the hallmark of intelligence, requiring strong generalization and abstraction skills. In a machine learning context, data-efficient methods are of high practical importance since data collection and annotation are prohibitively expensive in many domains. Thus, coordinated efforts to foster progress in this area emerged recently, e.g., in the form of dedicated workshops and competitions. Besides a common benchmark, measuring progress requires strong baselines. We present such a strong baseline for data-efficient image classification on the VIPriors challenge dataset, which is a sub-sampled version of ImageNet-1k with 100 images per class. We do not use any methods tailored to data-efficient classification but only standard models and techniques as well as common competition tricks and thorough hyper-parameter tuning. Our baseline achieves 69.7% accuracy on the VIPriors image classification dataset and outperforms 50% of submissions to the VIPriors 2021 challenge.

Wasim Ahmad, Maha Shadaydeh, Joachim Denzler

Estimating causal relations is vital in understanding the complex interactions in multivariate time series. Non-linear coupling of variables is one of the major challenges inaccurate estimation of cause-effect relations. In this paper, we propose to use deep autoregressive networks (DeepAR) in tandem with counterfactual analysis to infer nonlinear causal relations in multivariate time series. We extend the concept of Granger causality using probabilistic forecasting with DeepAR. Since deep networks can neither handle missing input nor out-of-distribution intervention, we propose to use the Knockoffs framework (Barberand Cand`es, 2015) for generating intervention variables and consequently counterfactual probabilistic forecasting. Knockoff samples are independent of their output given the observed variables and exchangeable with their counterpart variables without changing the underlying distribution of the data. We test our method on synthetic as well as real-world time series datasets. Overall our method outperforms the widely used vector autoregressive Granger causality and PCMCI in detecting nonlinear causal dependency in multivariate time series.

Violeta Teodora Trifunov, Maha Shadaydeh, Björn Barz et al.

There are numerous methods for detecting anomalies in time series, but that is only the first step to understanding them. We strive to exceed this by explaining those anomalies. Thus we develop a novel attribution scheme for multivariate time series relying on counterfactual reasoning. We aim to answer the counterfactual question of would the anomalous event have occurred if the subset of the involved variables had been more similarly distributed to the data outside of the anomalous interval. Specifically, we detect anomalous intervals using the Maximally Divergent Interval (MDI) algorithm, replace a subset of variables with their in-distribution values within the detected interval and observe if the interval has become less anomalous, by re-scoring it with MDI. We evaluate our method on multivariate temporal and spatio-temporal data and confirm the accuracy of our anomaly attribution of multiple well-understood extreme climate events such as heatwaves and hurricanes.

Lorenzo Brigato, Björn Barz, Luca Iocchi et al.

Data-efficient image classification using deep neural networks in settings, where only small amounts of labeled data are available, has been an active research area in the recent past. However, an objective comparison between published methods is difficult, since existing works use different datasets for evaluation and often compare against untuned baselines with default hyper-parameters. We design a benchmark for data-efficient image classification consisting of six diverse datasets spanning various domains (e.g., natural images, medical imagery, satellite data) and data types (RGB, grayscale, multispectral). Using this benchmark, we re-evaluate the standard cross-entropy baseline and eight methods for data-efficient deep learning published between 2017 and 2021 at renowned venues. For a fair and realistic comparison, we carefully tune the hyper-parameters of all methods on each dataset. Surprisingly, we find that tuning learning rate, weight decay, and batch size on a separate validation split results in a highly competitive baseline, which outperforms all but one specialized method and performs competitively to the remaining one.

Björn Barz, Joachim Denzler

We introduce a novel dataset for architectural style classification, consisting of 9,485 images of church buildings. Both images and style labels were sourced from Wikipedia. The dataset can serve as a benchmark for various research fields, as it combines numerous real-world challenges: fine-grained distinctions between classes based on subtle visual features, a comparatively small sample size, a highly imbalanced class distribution, a high variance of viewpoints, and a hierarchical organization of labels, where only some images are labeled at the most precise level. In addition, we provide 631 bounding box annotations of characteristic visual features for 139 churches from four major categories. These annotations can, for example, be useful for research on fine-grained classification, where additional expert knowledge about distinctive object parts is often available. Images and annotations are available at: https://doi.org/10.5281/zenodo.5166987