Konferenzbeitrag: h5-Index > 30
Refine
Document Type
Conference Type
- Konferenzartikel (24)
Language
- English (24)
Is part of the Bibliography
- yes (24)
Keywords
- Deep Leaning (4)
- Robustness (3)
- Festigkeit (2)
- Materialermüdung (2)
- Rissausbreitung (2)
- User Experience (2)
- Academic Advising (1)
- Additive Manufacturing (1)
- Adversarial Robustness (1)
- Bowtie antenna (1)
Institute
- Fakultät Elektrotechnik, Medizintechnik und Informatik (EMI) (ab 04/2019) (15)
- IMLA - Institute for Machine Learning and Analytics (10)
- Fakultät Wirtschaft (W) (6)
- ivESK - Institut für verlässliche Embedded Systems und Kommunikationselektronik (4)
- Fakultät Maschinenbau und Verfahrenstechnik (M+V) (2)
- Fakultät Medien (M) (ab 22.04.2021) (2)
- IUAS - Institute for Unmanned Aerial Systems (1)
Open Access
- Closed (15)
- Open Access (9)
- Diamond (5)
- Bronze (2)
- Grün (2)
Detecting Images Generated by Deep Diffusion Models using their Local Intrinsic Dimensionality
(2023)
Diffusion models recently have been successfully applied for the visual synthesis of strikingly realistic appearing images. This raises strong concerns about their potential for malicious purposes. In this paper, we propose using the lightweight multi Local Intrinsic Dimensionality (multiLID), which has been originally developed in context of the detection of adversarial examples, for the automatic detection of synthetic images and the identification of the according generator networks. In contrast to many existing detection approaches, which often only work for GAN-generated images, the proposed method provides close to perfect detection results in many realistic use cases. Extensive experiments on known and newly created datasets demonstrate that the proposed multiLID approach exhibits superiority in diffusion detection and model identification.Since the empirical evaluations of recent publications on the detection of generated images are often mainly focused on the "LSUN-Bedroom" dataset, we further establish a comprehensive benchmark for the detection of diffusion-generated images, including samples from several diffusion models with different image sizes.The code for our experiments is provided at https://github.com/deepfake-study/deepfake-multiLID.
Following their success in visual recognition tasks, Vision Transformers(ViTs) are being increasingly employed for image restoration. As a few recent works claim that ViTs for image classification also have better robustness properties, we investigate whether the improved adversarial robustness of ViTs extends to image restoration. We consider the recently proposed Restormer model, as well as NAFNet and the "Baseline network" which are both simplified versions of a Restormer. We use Projected Gradient Descent (PGD) and CosPGD for our robustness evaluation. Our experiments are performed on real-world images from the GoPro dataset for image deblurring. Our analysis indicates that contrary to as advocated by ViTs in image classification works, these models are highly susceptible to adversarial attacks. We attempt to find an easy fix and improve their robustness through adversarial training. While this yields a significant increase in robustness for Restormer, results on other networks are less promising. Interestingly, we find that the design choices in NAFNet and Baselines, which were based on iid performance, and not on robust generalization, seem to be at odds with the model robustness.
Public educational institutions are increasingly confronted with a decline in the number of applicants, which is why competition between colleges and universities is also intensifying. For this reason, it is important to position oneself as an institution in order to be perceived by the various target groups and to differentiate oneself from the competition. In this context, the brand and thus its perception and impact play a decisive role, especially in view of the desired communication of the institution's own values and its self-image, the brand identity. To this end, emotions serve as an approach to creating positive stimulation and brand loyalty.
The often-occurring short-term orders of manufactured products require a high machine availability. This requirement increases the importance of predictive maintenance solutions for bearings used in machines. There are, among others, hybrid solutions that rely on a physical model. For their usage, knowing the different degradation stages of bearings is essential. This research analyzes the underlying failure mechanisms of these stages theoretically and in a practical example of the well-known FEMTO dataset used for the IEEE PHM 2012 Data Challenge to provide this knowledge. In addition, it shows for which use cases the usage of low-frequency accelerometers is sufficient. The analysis provides that the degradation stages toward the end of the bearing life can also be detected with low-frequency accelerometers. Further, the importance of high-frequency accelerometers to detect bearing faults in early degradation stages is pointed out. These aspects have not been paid attention to by industry and research until now, despite providing a considerable cost-saving potential.
As cyber-attacks and functional safety requirements increase in Operational Technology (OT), implementing security measures becomes crucial. The IEC/IEEE 60802 draft standard addresses the security convergence in Time-Sensitive Networks (TSN) for industrial automation.We present the standard’s security architecture and its goals to establish end-to-end security with resource access authorization in OT systems. We compare the standard to our abstract technology-independent model for the management of cryptographic credentials during the lifecycles of OT systems. Additionally, we implemented the processes, mechanisms, and protocols needed for IEC/IEEE 60802 and extended the architecture with public key infrastructure (PKI) functionalities to support complete security management processes.
Investigation on Bowtie Antennas Operating at Very Low Frequencies for Ground Penetrating Radar
(2023)
The efficiency of Ground Penetrating Radar (GPR) systems significantly depends on the antenna performance as the signal has to propagate through lossy and inhomogeneous media. GPR antennas should have a low operating frequency for greater penetration depth, high gain and efficiency to increase the receiving power and should be compact and lightweight for ease of GPR surveying. In this paper, two different designs of Bowtie antennas operating at very low frequencies are proposed and analyzed.
It is common practice to apply padding prior to convolution operations to preserve the resolution of feature-maps in Convolutional Neural Networks (CNN). While many alternatives exist, this is often achieved by adding a border of zeros around the inputs. In this work, we show that adversarial attacks often result in perturbation anomalies at the image boundaries, which are the areas where padding is used. Consequently, we aim to provide an analysis of the interplay between padding and adversarial attacks and seek an answer to the question of how different padding modes (or their absence) affect adversarial robustness in various scenarios.
Seismic data processing relies on multiples attenuation to improve inversion and interpretation. Radon-based algorithms are often used for multiples and primaries discrimination. Deep learning, based on convolutional neural networks (CNNs), has shown encouraging applications for demultiple that could mitigate Radon-based challenges. In this work, we investigate new strategies to train a CNN for multiples removal based on different loss functions. We propose combined primaries and multiples labels in the loss for training a CNN to predict primaries, multiples, or both simultaneously. Moreover, we investigate two distinctive training methods for all the strategies: UNet based on minimum absolute error (L1) training, and adversarial training (GAN-UNet). We test the trained models with the different strategies and methods on 400 synthetic data. We found that training to predict multiples, including the primaries …
Seismic data processing involves techniques to deal with undesired effects that occur during acquisition and pre-processing. These effects mainly comprise coherent artefacts such as multiples, non-coherent signals such as electrical noise, and loss of signal information at the receivers that leads to incomplete traces. In this work, we employ a generative solution, since it can explicitly model complex data distributions and hence, yield to a better decision-making process. In particular, we introduce diffusion models for multiple removal. To that end, we run experiments on synthetic and on real data, and we compare the deep diffusion performance with standard algorithms. We believe that our pioneer study not only demonstrates the capability of diffusion models, but also opens the door to future research to integrate generative models in seismic workflows.
Neural networks have a number of shortcomings. Amongst the severest ones is the sensitivity to distribution shifts which allows models to be easily fooled into wrong predictions by small perturbations to inputs that are often imperceivable to humans and do not have to carry semantic meaning. Adversarial training poses a partial solution to address this issue by training models on worst-case perturbations. Yet, recent work has also pointed out that the reasoning in neural networks is different from humans. Humans identify objects by shape, while neural nets mainly employ texture cues. Exemplarily, a model trained on photographs will likely fail to generalize to datasets containing sketches. Interestingly, it was also shown that adversarial training seems to favorably increase the shift toward shape bias. In this work, we revisit this observation and provide an extensive analysis of this effect on various architectures, the common L_2-and L_-training, and Transformer-based models. Further, we provide a possible explanation for this phenomenon from a frequency perspective.
The present paper addresses the research question: What recommendations for action and potential adjustments should an online magazine for beauty and fashion implement in order to make affiliate articles in these sections even more appealing to the target group and provide added value for them?
To be able to answer this research question, three hypotheses were defined and tested with using qualitative and quantitative research. The qualitative research consisted of user experience testings, where four affiliate articles in the fields of beauty and fashion were tested with 13 participants. The quantitative research involved collecting, analyzing and evaluating data from the four affiliate articles conducted with the company's real-life target group. Based on these results, recommendations for action were derived, which should not only improve the quality of the content in the future, but also increase the efficiency of the implementation of those articles.
In order to attract new students, German universities must provide quick and easy access to relevant information. A chatbot can help increase the efficiency in academic advising for prospective students. In this study we evaluate the acceptance and effects of chatbots in German student-university communication. We conducted a qualitative UX-Study with the chatbot prototype of Offenburg University of Applied Sciences (HSO), in order to determine which features are particularly relevant and which requirements are made by the users. The results show that acceptance increases if the chatbot offers quick and adequate assistance, furthermore, our participants preferred an informal communication style and valued friendly and helpful personality traits for chatbots.
Turbocharger housings in internal combustion engines are subjected to severe mechanical and thermal cyclic loads throughout their life-time or during engine testing. The combination of thermal transients and mechanical load cycling results in a complex evolution of damage, leading to thermo-mechanical fatigue (TMF) of the material. For the computational TMF life assessment of high temperature components, the DTMF model can provide reliable TMF life predictions. The model is based on a short fatigue crack growth law and uses local finite-element (FE) results to predict the number of cycles to failure for a technical crack. In engine applications, it is nowadays often acceptable to have short cracks as long as they do not propagate and cause loss of function of the component. Thus, it is necessary to predict not only potential crack locations and the corresponding number of cycles for a technical crack, but also to determine subsequent crack growth or even a possible crack arrest. In this work, a method is proposed that allows the simulation of TMF crack growth in high temperature components using FE simulations and non-linear fracture mechanics (NLFM).
A NLFM based crack growth simulation method is described. This method starts with the FE analysis of a component. In this paper, the method is demonstrated for an automotive turbocharger housing subjected to TMF loading. A transient elastic-viscoplastic FE analysis is used to simulate four heating and cooling cycles of an engine test. The stresses, inelastic strains, and temperature histories from the FEA are then used to perform TMF life predictions using the standard DTMF model. The crack position and the crack plane of critical hotspots are then identified. Simulated cracks are inserted at the hotspots. For the model demonstrated, cracks were inserted at two hotspot locations. The ΔJ integral is computed as a fracture mechanics parameter at each point along the crack-front, and the crack extension of each point is then evaluated, allowing the crack to grow iteratively. The paper concludes with a comparison of the crack growth curves for both hotspots with experimental results.
Cast aluminum cylinder blocks are frequently used in gasoline and diesel internal combustion engines because of their light-weight advantage. However, the disadvantage of aluminum alloys is their relatively low strength and fatigue resistance which make aluminum blocks prone to fatigue cracking. Engine blocks must withstand a combination of low-cycle fatigue (LCF) thermal loads and high-cycle fatigue (HCF) combustion and dynamic loads. Reliable computational methods are needed that allow for accurate fatigue assessment of cylinder blocks under this combined loading. In several publications, the mechanism-based thermomechanical fatigue (TMF) damage model DTMF describing the growth of short fatigue cracks has been extended to include the effect of both LCF thermal loads and superimposed HCF loadings. This approach is applied to the finite life fatigue assessment of an aluminum cylinder block. The required material properties related to LCF are determined from uniaxial LCF tests. The additional material properties required for the assessment of superimposed HCF are obtained from the literature for similar materials. The predictions of the model agree well with engine dyno test results. Finally, some improvements to the current process are discussed.
Complex tourism products with intangible service components are difficult to explain to potential customers. This research elaborates the use of virtual reality (VR) in the field of shore excursions. A theoretical research model based on the technology acceptance model was developed, and hypotheses were proposed. Cruise passengers were invited to test 360° excursion images on a landing page. Data was collected using an online questionnaire. Finally, data was analyzed using the PLS-SEM method. The results provide theoretical implications on technology acceptance model (TAM) research in the field of cruise tourism. Furthermore, the results and implications indicate the potential of virtual 360° shore excursion presentations for the cruise industry.
Additive manufacturing enables the production of lightweight and resilient components with extensive design freedom. In the low-cost sector, material extrusion (e.g. Fused Deposition Modeling - FDM) has been the main method used to date. Thus, robust 3D printers and inexpensive 3D materials (polymer filaments) can be used. However, the printing times for FDM are very long and the quality of the dimensions and surfaces is limited. Recently, new processes from the field of Vat polymerization have entered the market. For example, masked stereolithography (mSLA) offers a significant improvement in component quality and build speed through the use of resins and large-area curing at still reasonable costs. Currently, there is only limited knowledge available on the optimal design of components using this young process. In this contribution, design guidelines are developed to determine the possibilities and limitations of mSLA from a design point of view. For this purpose, a number of test geometries are designed and investigated to obtain systematic insights into important design features, such as wall thickness, grooves and holes. In addition, typical problems in additive manufacturing, such as the design of overhangs and fits or the hollowing of components, are investigated. The evaluation of practical 3D printing tests thus provides important parameters that can be transferred to design guidelines of components for additive manufacturing using mSLA.
In this paper, we study the runtime performance of symmetric cryptographic algorithms on an embedded ARM Cortex-M4 platform. Symmetric cryptographic algorithms can serve to protect the integrity and optionally, if supported by the algorithm, the confidentiality of data. A broad range of well-established algorithms exists, where the different algorithms typically have different properties and come with different computational complexity. On deeply embedded systems, the overhead imposed by cryptographic operations may be significant. We execute the algorithms AES-GCM, ChaCha20-Poly1305, HMAC-SHA256, KMAC, and SipHash on an STM32 embedded microcontroller and benchmark the execution times of the algorithms as a function of the input lengths.
In recent years, the topic of embedded machine learning has become very popular in AI research. With the help of various compression techniques such as pruning, quantization and others compression techniques, it became possible to run neural networks on embedded devices. These techniques have opened up a whole new application area for machine learning. They range from smart products such as voice assistants to smart sensors that are needed in robotics. Despite the achievements in embedded machine learning, efficient algorithms for training neural networks in constrained domains are still lacking. Training on embedded devices will open up further fields of applications. Efficient training algorithms would enable federated learning on embedded devices, in which the data remains where it was collected, or retraining of neural networks in different domains. In this paper, we summarize techniques that make training on embedded devices possible. We first describe the need and requirements for such algorithms. Then we examine existing techniques that address training in resource-constrained environments as well as techniques that are also suitable for training on embedded devices, such as incremental learning. At the end, we also discuss which problems and open questions still need to be solved in these areas.
Currently, many theoretical as well as practically relevant questions towards the transferability and robustness of Convolutional Neural Networks (CNNs) remain unsolved. While ongoing research efforts are engaging these problems from various angles, in most computer vision related cases these approaches can be generalized to investigations of the effects of distribution shifts in image data. In this context, we propose to study the shifts in the learned weights of trained CNN models. Here we focus on the properties of the distributions of dominantly used 3×3 convolution filter kernels. We collected and publicly provide a dataset with over 1.4 billion filters from hundreds of trained CNNs, using a wide range of datasets, architectures, and vision tasks. In a first use case of the proposed dataset, we can show highly relevant properties of many publicly available pre-trained models for practical applications: I) We analyze distribution shifts (or the lack thereof) between trained filters along different axes of meta-parameters, like visual category of the dataset, task, architecture, or layer depth. Based on these results, we conclude that model pre-training can succeed on arbitrary datasets if they meet size and variance conditions. II) We show that many pre-trained models contain degenerated filters which make them less robust and less suitable for fine-tuning on target applications. Data & Project website: https://github.com/paulgavrikov/cnn-filter-db.
Deep learning models are intrinsically sensitive to distribution shifts in the input data. In particular, small, barely perceivable perturbations to the input data can force models to make wrong predictions with high confidence. An common defense mechanism is regularization through adversarial training which injects worst-case perturbations back into training to strengthen the decision boundaries, and to reduce overfitting. In this context, we perform an investigation of 3 × 3 convolution filters that form in adversarially- trained models. Filters are extracted from 71 public models of the ℓ ∞ -RobustBench CIFAR-10/100 and ImageNet1k leaderboard and compared to filters extracted from models built on the same architectures but trained without robust regularization. We observe that adversarially-robust models appear to form more diverse, less sparse, and more orthogonal convolution filters than their normal counterparts. The largest differences between robust and normal models are found in the deepest layers, and the very first convolution layer, which consistently and predominantly forms filters that can partially eliminate perturbations, irrespective of the architecture.