Refine
Document Type
- Conference Proceeding (63) (remove)
Conference Type
- Konferenzartikel (42)
- Konferenz-Abstract (15)
- Sonstiges (5)
- Konferenz-Poster (1)
Language
- English (63) (remove)
Is part of the Bibliography
- yes (63)
Keywords
- RoboCup (12)
- Machine Learning (8)
- Deep Leaning (4)
- Generative Adversarial Network (3)
- Robustness (3)
- Computer Vision (2)
- Heart rhythm model (2)
- Modeling and simulation (2)
- Neural networks (2)
- Stability (2)
Institute
- Fakultät Elektrotechnik, Medizintechnik und Informatik (EMI) (ab 04/2019) (63) (remove)
Open Access
- Open Access (63) (remove)
Team description papers of magmaOffenburg are incremental in the sense that each year we address a different topic of our team and the tools around our team. In this year’s team description paper we focus on the architecture of the software. It is a main factor for being able to keep the code maintainable even after 15 years of development. We also describe how we make sure that the code follows this architecture.
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 this paper, we propose a unified approach for network pruning and one-shot neural architecture search (NAS) via group sparsity. We first show that group sparsity via the recent Proximal Stochastic Gradient Descent (ProxSGD) algorithm achieves new state-of-the-art results for filter pruning. Then, we extend this approach to operation pruning, directly yielding a gradient-based NAS method based on group sparsity. Compared to existing gradient-based algorithms such as DARTS, the advantages of this new group sparsity approach are threefold. Firstly, instead of a costly bilevel optimization problem, we formulate the NAS problem as a single-level optimization problem, which can be optimally and efficiently solved using ProxSGD with convergence guarantees. Secondly, due to the operation-level sparsity, discretizing the network architecture by pruning less important operations can be safely done without any performance degradation. Thirdly, the proposed approach finds architectures that are both stable and well-performing on a variety of search spaces and datasets.
We demonstrate how to exploit group sparsity in order to bridge the areas of network pruning and neural architecture search (NAS). This results in a new one-shot NAS optimizer that casts the problem as a single-level optimization problem and does not suffer any performance degradation from discretizating the architecture.
The use of artificial intelligence continues to impact a broad variety of domains, application areas, and people. However, interpretability, understandability, responsibility, accountability, and fairness of the algorithms' results - all crucial for increasing humans' trust into the systems - are still largely missing. The purpose of this seminar is to understand how these components factor into the holistic view of trust. Further, this seminar seeks to identify design guidelines and best practices for how to build interactive visualization systems to calibrate trust.
Sweaty has already participated several times in RoboCup soccer competitions (Adult Size). Now the work is focused on stabilizing the gait. Moreover, we would like to overcome the constraints of a ZMP-algorithm that has a horizontal footplate as precondition for the simplification of the equations. In addition we would like to switch between impedance and position control with a fuzzy-like algorithm that might help to minimize jerks when Sweaty’s feet touch the ground.
Sweaty has already participated four times in RoboCup soccer competitions (Adult Size) and came second three times. While 2016 Sweaty needed a lot of luck to be finalist, 2017 Sweaty was a serious adversary in the preliminary rounds. In 2018 Sweaty showed up in the final with some lack of experience and room for improvements, but not without any chance. This paper describes the intended improvements of the humanoid adult size robot Sweaty in order to qualify for the RoboCup 2019 adult size competition.
Generative adversarial networks (GANs) provide state-of-the-art results in image generation. However, despite being so powerful, they still remain very challenging to train. This is in particular caused by their highly non-convex optimization space leading to a number of instabilities. Among them, mode collapse stands out as one of the most daunting ones. This undesirable event occurs when the model can only fit a few modes of the data distribution, while ignoring the majority of them. In this work, we combat mode collapse using second-order gradient information. To do so, we analyse the loss surface through its Hessian eigenvalues, and show that mode collapse is related to the convergence towards sharp minima. In particular, we observe how the eigenvalues of the are directly correlated with the occurrence of mode collapse. Finally, motivated by these findings, we design a new optimization algorithm called nudged-Adam (NuGAN) that uses spectral information to overcome mode collapse, leading to empirically more stable convergence properties.
Generative adversarial networks are the state of the art approach towards learned synthetic image generation. Although early successes were mostly unsupervised, bit by bit, this trend has been superseded by approaches based on labelled data. These supervised methods allow a much finer-grained control of the output image, offering more flexibility and stability. Nevertheless, the main drawback of such models is the necessity of annotated data. In this work, we introduce an novel framework that benefits from two popular learning techniques, adversarial training and representation learning, and takes a step towards unsupervised conditional GANs. In particular, our approach exploits the structure of a latent space (learned by the representation learning) and employs it to condition the generative model. In this way, we break the traditional dependency between condition and label, substituting the latter by unsupervised features coming from the latent space. Finally, we show that this new technique is able to produce samples on demand keeping the quality of its supervised counterpart.
Facial image manipulation is a generation task where the output face is shifted towards an intended target direction in terms of facial attribute and styles. Recent works have achieved great success in various editing techniques such as style transfer and attribute translation. However, current approaches are either focusing on pure style transfer, or on the translation of predefined sets of attributes with restricted interactivity. To address this issue, we propose FacialGAN, a novel framework enabling simultaneous rich style transfers and interactive facial attributes manipulation. While preserving the identity of a source image, we transfer the diverse styles of a target image to the source image. We then incorporate the geometry information of a segmentation mask to provide a fine-grained manipulation of facial attributes. Finally, a multi-objective learning strategy is introduced to optimize the loss of each specific tasks. Experiments on the CelebA-HQ dataset, with CelebAMask-HQ as semantic mask labels, show our model’s capacity in producing visually compelling results in style transfer, attribute manipulation, diversity and face verification. For reproducibility, we provide an interactive open-source tool to perform facial manipulations, and the Pytorch implementation of the model.
A fundamental and still largely unsolved question in the context of Generative Adversarial Networks is whether they are truly able to capture the real data distribution and, consequently, to sample from it. In particular, the multidimensional nature of image distributions leads to a complex evaluation of the diversity of GAN distributions. Existing approaches provide only a partial understanding of this issue, leaving the question unanswered. In this work, we introduce a loop-training scheme for the systematic investigation of observable shifts between the distributions of real training data and GAN generated data. Additionally, we introduce several bounded measures for distribution shifts, which are both easy to compute and to interpret. Overall, the combination of these methods allows an explorative investigation of innate limitations of current GAN algorithms. Our experiments on different data-sets and multiple state-of-the-art GAN architectures show large shifts between input and output distributions, showing that existing theoretical guarantees towards the convergence of output distributions appear not to be holding in practice.
Generative convolutional deep neural networks, e.g. popular GAN architectures, are relying on convolution based up-sampling methods to produce non-scalar outputs like images or video sequences. In this paper, we show that common up-sampling methods, i.e. known as up-convolution or transposed convolution, are causing the inability of such models to reproduce spectral distributions of natural training data correctly. This effect is independent of the underlying architecture and we show that it can be used to easily detect generated data like deepfakes with up to 100% accuracy on public benchmarks. To overcome this drawback of current generative models, we propose to add a novel spectral regularization term to the training optimization objective. We show that this approach not only allows to train spectral consistent GANs that are avoiding high frequency errors. Also, we show that a correct approximation of the frequency spectrum has positive effects on the training stability and output quality of generative networks.
In this preliminary report, we present a simple but very effective technique to stabilize the training of CNN based GANs. Motivated by recently published methods using frequency decomposition of convolutions (eg Octave Convolutions), we propose a novel convolution scheme to stabilize the training and reduce the likelihood of a mode collapse. The basic idea of our approach is to split convolutional filters into additive high and low frequency parts, while shifting weight updates from low to high during the training. Intuitively, this method forces GANs to learn low frequency coarse image structures before descending into fine (high frequency) details. Our approach is orthogonal and complementary to existing stabilization methods and can simply plugged into any CNN based GAN architecture. First experiments on the CelebA dataset show the effectiveness of the proposed method.
Spatially Distributed Wireless Networks (SDWN) are one of the basic technologies for the Internet of Things (IoT) and (Industrial) Internet of Things (IIoT) applications. These SDWN for many of these applications has strict requirements such as low cost, simple installation and operations, and high potential flexibility and mobility. Among the different Narrowband Wireless Wide Area Networking (NBWWAN) technologies, which are introduced to address these categories of wireless networking requirements, Narrowband Internet of Things (NB-IoT) is getting more traction due to attractive system parameters, energy-saving mode of operation with low data rates and bandwidth, and its applicability in 5G use cases. Since several technologies are available and because the underlying use cases come with various requirements, it is essential to perform a systematic comparative analysis of competing technologies to choose the right technology. It is also important to perform testing during different phases of the system development life cycle. This paper describes the systematic test environment for automated testing of radio communication and systematic measurements of the performance of NB-IoT.
Electrode modelling and simulation of diagnostic and pulmonary vein isolation in atrial fibrillation
(2022)
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.
Recent work has investigated the distributions of learned convolution filters through a large-scale study containing hundreds of heterogeneous image models. Surprisingly, on average, the distributions only show minor drifts in comparisons of various studied dimensions including the learned task, image domain, or dataset. However, among the studied image domains, medical imaging models appeared to show significant outliers through "spikey" distributions, and, therefore, learn clusters of highly specific filters different from other domains. Following this observation, we study the collected medical imaging models in more detail. We show that instead of fundamental differences, the outliers are due to specific processing in some architectures. Quite the contrary, for standardized architectures, we find that models trained on medical data do not significantly differ in their filter distributions from similar architectures trained on data from other domains. Our conclusions reinforce previous hypotheses stating that pre-training of imaging models can be done with any kind of diverse image data.
An Empirical Investigation of Model-to-Model Distribution Shifts in Trained Convolutional Filters
(2021)
We present first empirical results from our ongoing investigation of distribution shifts in image data used for various computer vision tasks. Instead of analyzing the original training and test data, we propose to study shifts in the learned weights of trained models. In this work, we focus on the properties of the distributions of dominantly used 3x3 convolution filter kernels. We collected and publicly provide a data set with over half a billion filters from hundreds of trained CNNs, using a wide range of data sets, architectures, and vision tasks. Our analysis shows interesting distribution shifts (or the lack thereof) between trained filters along different axes of meta-parameters, like data type, task, architecture, or layer depth. We argue, that the observed properties are a valuable source for further investigation into a better understanding of the impact of shifts in the input data to the generalization abilities of CNN models and novel methods for more robust transfer-learning in this domain.
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 importance of machine learning has been increasing dramatically for years. From assistance systems to production optimisation to support the health sector, almost every area of daily life and industry comes into contact with machine learning. Besides all the benefits that ML brings, the lack of transparency and the difficulty in creating traceability pose major risks. While there are solutions that make the training of machine learning models more transparent, traceability is still a major challenge. Ensuring the identity of a model is another challenge. Unnoticed modification of a model is also a danger when using ML. One solution is to create an ML birth certificate and an ML family tree secured by blockchain technology. Important information about training and changes to the model through retraining can be stored in a blockchain and accessed by any user to create more security and traceability about an ML model.
Despite the success of convolutional neural networks (CNNs) in many academic benchmarks for computer vision tasks, their application in the real-world is still facing fundamental challenges. One of these open problems is the inherent lack of robustness, unveiled by the striking effectiveness of adversarial attacks. Current attack methods are able to manipulate the network's prediction by adding specific but small amounts of noise to the input. In turn, adversarial training (AT) aims to achieve robustness against such attacks and ideally a better model generalization ability by including adversarial samples in the trainingset. However, an in-depth analysis of the resulting robust models beyond adversarial robustness is still pending. In this paper, we empirically analyze a variety of adversarially trained models that achieve high robust accuracies when facing state-of-the-art attacks and we show that AT has an interesting side-effect: it leads to models that are significantly less overconfident with their decisions, even on clean data than non-robust models. Further, our analysis of robust models shows that not only AT but also the model's building blocks (like activation functions and pooling) have a strong influence on the models' prediction confidences. Data & Project website: https://github.com/GeJulia/robustness_confidences_evaluation
Despite the success of convolutional neural networks (CNNs) in many academic benchmarks for computer vision tasks, their application in the real-world is still facing fundamental challenges. One of these open problems is the inherent lack of robustness, unveiled by the striking effectiveness of adversarial attacks. Adversarial training (AT) is often considered as a remedy to train more robust networks. In this paper, we empirically analyze a variety of adversarially trained models that achieve high robust accuracies when facing state-of-the-art attacks and we show that AT has an interesting side-effect: it leads to models that are significantly less overconfident with their decisions even on clean data than non-robust models. Further, our analysis of robust models shows that not only AT but also the model's building blocks (like activation functions and pooling) have a strong influence on the models' prediction confidences.
Over the last years, Convolutional Neural Networks (CNNs) have been the dominating neural architecture in a wide range of computer vision tasks. From an image and signal processing point of view, this success might be a bit surprising as the inherent spatial pyramid design of most CNNs is apparently violating basic signal processing laws, i.e. Sampling Theorem in their down-sampling operations. However, since poor sampling appeared not to affect model accuracy, this issue has been broadly neglected until model robustness started to receive more attention. Recent work in the context of adversarial attacks and distribution shifts, showed after all, that there is a strong correlation between the vulnerability of CNNs and aliasing artifacts induced by poor down-sampling operations. This paper builds on these findings and introduces an aliasing free down-sampling operation which can easily be plugged into any CNN architecture: FrequencyLowCut pooling. Our experiments show, that in combination with simple and Fast Gradient Sign Method (FGSM) adversarial training, our hyper-parameter free operator substantially improves model robustness and avoids catastrophic overfitting. Our code is available at https://github.com/GeJulia/flc_pooling
Many commonly well-performing convolutional neural network models have shown to be susceptible to input data perturbations, indicating a low model robustness. Adversarial attacks are thereby specifically optimized to reveal model weaknesses, by generating small, barely perceivable image perturbations that flip the model prediction. Robustness against attacks can be gained for example by using adversarial examples during training, which effectively reduces the measurable model attackability. In contrast, research on analyzing the source of a model’s vulnerability is scarce. In this paper, we analyze adversarially trained, robust models in the context of a specifically suspicious network operation, the downsampling layer, and provide evidence that robust models have learned to downsample more accurately and suffer significantly less from aliasing than baseline models.
Aerosol particles play an important role in the climate system by absorbing and scattering radiation and influencing cloud properties. They are also one of the biggest sources of uncertainty for climate modeling. Many climate models do not include aerosols in sufficient detail. In order to achieve higher accuracy, aerosol microphysical properties and processes have to be accounted for. This is done in the ECHAM-HAM global climate aerosol model using the M7 microphysics model, but increased computational costs make it very expensive to run at higher resolutions or for a longer time. We aim to use machine learning to approximate the microphysics model at sufficient accuracy and reduce the computational cost by being fast at inference time. The original M7 model is used to generate data of input-output pairs to train a neural network on it. By using a special logarithmic transform we are able to learn the variables tendencies achieving an average score of . On a GPU we achieve a speed-up of 120 compared to the original model.
The present work ties in with the problem of bicycle road assessment that is currently done using expensive special measuring vehicles. Our alternative approach for road condition assessment is to mount a sensor device on a bicycle which sends accelerometer and gyroscope data via WiFi to a classification server. There, a prediction model determines road type and condition based on the sensor data. For the classification task, we compare different machine learning methods with each other, whereby validation accuracies of 99% can be achieved with deep residual networks such as InceptionTime. The main contribution of this work with respect to comparable work is that we achieve excellent accuracies on a realistic dataset classifying road conditions into nine distinct classes that are highly relevant for practice.
Sweaty has already participated several times in RoboCup soccer competitions (Adult Size). Now the work is focused coordinating the play of two robots. Moreover, we are working on stabilizing the gait by adding additional sensor information. An ongoing work is the optimization of the control strategy by balancing between impedance and position control. By minimizing the jerk, gait and overall gameplay should improve significantly.
Multiple Object Tracking (MOT) is a long-standing task in computer vision. Current approaches based on the tracking by detection paradigm either require some sort of domain knowledge or supervision to associate data correctly into tracks. In this work, we present a self-supervised multiple object tracking approach based on visual features and minimum cost lifted multicuts. Our method is based on straight-forward spatio-temporal cues that can be extracted from neighboring frames in an image sequences without supervision. Clustering based on these cues enables us to learn the required appearance invariances for the tracking task at hand and train an AutoEncoder to generate suitable latent representations. Thus, the resulting latent representations can serve as robust appearance cues for tracking even over large temporal distances where no reliable spatio-temporal features can be extracted. We show that, despite being trained without using the provided annotations, our model provides competitive results on the challenging MOT Benchmark for pedestrian tracking.
In this work, we evaluate two different image clustering objectives, k-means clustering and correlation clustering, in the context of Triplet Loss induced feature space embeddings. Specifically, we train a convolutional neural network to learn discriminative features by optimizing two popular versions of the Triplet Loss in order to study their clustering properties under the assumption of noisy labels. Additionally, we propose a new, simple Triplet Loss formulation, which shows desirable properties with respect to formal clustering objectives and outperforms the existing methods. We evaluate all three Triplet loss formulations for K-means and correlation clustering on the CIFAR-10 image classification dataset.
Estimating the Robustness of Classification Models by the Structure of the Learned Feature-Space
(2022)
Over the last decade, the development of deep image classification networks has mostly been driven by the search for the best performance in terms of classification accuracy on standardized benchmarks like ImageNet. More recently, this focus has been expanded by the notion of model robustness, \ie the generalization abilities of models towards previously unseen changes in the data distribution. While new benchmarks, like ImageNet-C, have been introduced to measure robustness properties, we argue that fixed testsets are only able to capture a small portion of possible data variations and are thus limited and prone to generate new overfitted solutions. To overcome these drawbacks, we suggest to estimate the robustness of a model directly from the structure of its learned feature-space. We introduce robustness indicators which are obtained via unsupervised clustering of latent representations from a trained classifier and show very high correlations to the model performance on corrupted test data.
Engineering, construction and operation of complex machines involves a wide range of complicated, simultaneous tasks, which potentially could be automated. In this work, we focus on perception tasks in such systems, investigating deep learning approaches for multi-task transfer learning with limited training data. We show an approach that takes advantage of a technical systems’ focus on selected objects and their properties. We create focused representations and simultaneously solve joint objectives in a system through multi-task learning with convolutional autoencoders. The focused representations are used as a starting point for the data-saving solution of the additional tasks. The efficiency of this approach is demonstrated using images and tasks of an autonomous circular crane with a grapple.
This paper describes the concept and some results of the project "Menschen Lernen Maschinelles Lernen" (Humans Learn Machine Learning, ML2) of the University of Applied Sciences Offenburg. It brings together students of different courses of study and practitioners from companies on the subject of Machine Learning. A mixture of blended learning and practical projects ensures a tight coupling of machine learning theory and application. The paper details the phases of ML2 and mentions two successful example projects.
Active participation of industrial enterprises in electricity markets - a generic modeling approach
(2021)
Industrial enterprises represent a significant portion of electricity consumers with the potential of providing demand-side energy flexibility from their production processes and on-site energy assets. Methods are needed for the active and profitable participation of such enterprises in the electricity markets especially with variable prices, where the energy flexibility available in their manufacturing, utility and energy systems can be assessed and quantified. This paper presents a generic model library equipped with optimal control for energy flexibility purposes. The components in the model library represent the different technical units of an industrial enterprise on material, media, and energy flow levels with their process constraints. The paper also presents a case study simulation of a steel-powder manufacturing plant using the model library. Its energy flexibility was assessed when the plant procured its electrical energy at fixed and variable electricity prices. In the simulated case study, flexibility use at dynamic prices resulted in a 6% cost reduction compared to a fixed-price scenario, with battery storage and the manufacturing system making the largest contributions to flexibility.
In this paper, we describe a first publicly available fine-grained product recognition dataset based on leaflet images. Using advertisement leaflets, collected over several years from different European retailers, we provide a total of 41.6k manually annotated product images in 832 classes. Further, we investigate three different approaches for this fine-grained product classification task, Classification by Image, by Text, as well as by Image and Text. The approach "Classification by Text" uses the text extracted directly from the leaflet product images. We show, that the combination of image and text as input improves the classification of visual difficult to distinguish products. The final model leads to an accuracy of 96.4% with a Top-3 score of 99.2%. We release our code at https://github.com/ladwigd/Leaflet-Product-Classification.
Disturbances of the cardiac conduction system causing reentry mechanisms above the atrioventricular (AV) node are induced by at least one accessory pathway with different conducting properties and refractory periods. This work aims to further develop the already existing and continuously expanding Offenburg heart rhythm model to visualise the most common supraventricular reentry tachycardias to provide a better understanding of the cause of the respective reentry mechanism.
Recently, adversarial attacks on image classification networks by the AutoAttack (Croce and Hein, 2020b) framework have drawn a lot of attention. While AutoAttack has shown a very high attack success rate, most defense approaches are focusing on network hardening and robustness enhancements, like adversarial training. This way, the currently best-reported method can withstand about 66% of adversarial examples on CIFAR10. In this paper, we investigate the spatial and frequency domain properties of AutoAttack and propose an alternative defense. Instead of hardening a network, we detect adversarial attacks during inference, rejecting manipulated inputs. Based on a rather simple and fast analysis in the frequency domain, we introduce two different detection algorithms. First, a black box detector that only operates on the input images and achieves a detection accuracy of 100% on the AutoAttack CIFAR10 benchmark and 99.3% on ImageNet, for epsilon = 8/255 in both cases. Second, a whitebox detector using an analysis of CNN feature maps, leading to a detection rate of also 100% and 98.7% on the same benchmarks.
Convolutional neural networks (CNN) define the state-of-the-art solution on many perceptual tasks. However, current CNN approaches largely remain vulnerable against adversarial perturbations of the input that have been crafted specifically to fool the system while being quasi-imperceptible to the human eye. In recent years, various approaches have been proposed to defend CNNs against such attacks, for example by model hardening or by adding explicit defence mechanisms. Thereby, a small “detector” is included in the network and trained on the binary classification task of distinguishing genuine data from data containing adversarial perturbations. In this work, we propose a simple and light-weight detector, which leverages recent findings on the relation between networks’ local intrinsic dimensionality (LID) and adversarial attacks. Based on a re-interpretation of the LID measure and several simple adaptations, we surpass the state-of-the-art on adversarial detection by a significant m argin and reach almost perfect results in terms of F1-score for several networks and datasets. Sources available at: https://github.com/adverML/multiLID
Recently, RobustBench (Croce et al. 2020) has become a widely recognized benchmark for the adversarial robustness of image
classification networks. In it’s most commonly reported sub-task, RobustBench evaluates and ranks the adversarial robustness of trained neural networks on CIFAR10 under AutoAttack (Croce and Hein 2020b) with l∞ perturbations limited to ϵ = 8/255. With leading scores of the currently best performing models of around 60% of the baseline, it is fair to characterize this benchmark to be quite challenging. Despite it’s general acceptance in recent literature, we aim to foster discussion about the suitability of RobustBench as a key indicator for robustness which could be generalized to practical applications. Our line of argumentation against this is two-fold and supported by excessive experiments presented in this paper: We argue that I) the alternation of data by AutoAttack with l∞, ϵ = 8/255 is unrealistically strong, resulting in close to perfect detection rates of adversarial samples even by simple detection algorithms and human observers.
We also show that other attack methods are much harder to detect while achieving similar success rates. II) That results on low resolution data sets like CIFAR10 do not generalize well to higher resolution images as gradient based attacks appear to become even more detectable with increasing resolutions.
Harnessing the overall benefits of the latest advancements in artificial intelligence (AI) requires the extensive collaboration of academia and industry. These collaborations promote innovation and growth while enforcing the practical usefulness of newer technologies in real life. The purpose of this article is to outline the challenges faced during cross-collaboration between academia and industry. These challenges are also inspected with the help of an ongoing project titled “Quality Assurance of Machine Learning Applications” (Q-AMeLiA), in which three universities cooperate with five industry partners to make the product risk of AI-based products visible. Further, we discuss the hurdles and the key challenges in machine learning (ML) technology transformation from academia to industry based on robustness, simplicity, and safety. These challenges are an outcome of the lack of common standards, metrics, and missing regulatory considerations when state-of-the-art (SOTA) technology is developed in academia. The use of biased datasets involves ethical concerns that might lead to unfair outcomes when the ML model is deployed in production. The advancement of AI in small and medium sized enterprises (SMEs) requires more in terms of common tandardization of concepts rather than algorithm breakthroughs. In this paper, in addition to the general challenges, we also discuss domain specific barriers for five different domains i.e., object detection, hardware benchmarking, continual learning, action recognition, and industrial process automation, and highlight the steps necessary for successfully managing the cross-sectoral collaborations between academia and industry.
Apache Hadoop is a well-known open-source framework for storing and processing huge amounts of data. This paper shows the usage of the framework within a project of the university in cooperation with a semiconductor company. The goal of this project was to supplement the existing data landscape by the facilities of storing and analyzing the data on a new Apache Hadoop based platform.
Artificial intelligence (AI), and in particular machine learning algorithms, are of increasing importance in many application areas but interpretability and understandability as well as responsibility, accountability, and fairness of the algorithms' results, all crucial for increasing the humans' trust into the systems, are still largely missing. Big industrial players, including Google, Microsoft, and Apple, have become aware of this gap and recently published their own guidelines for the use of AI in order to promote fairness, trust, interpretability, and other goals. Interactive visualization is one of the technologies that may help to increase trust in AI systems. During the seminar, we discussed the requirements for trustworthy AI systems as well as the technological possibilities provided by interactive visualizations to increase human trust in AI.
Background: A disturbed synchronization of the ventricular contraction can cause a highly developed systolic heart failure in affected patients, which can often be explained by a diseased left bundle branch block (LBBB). If medication remains unresponsive, the concerned patients will be treated with a cardiac resynchronization therapy (CRT) system. The aim of this study was to integrate His bundle pacing into the Offenburg heart rhythm model in order to visualize the electrical pacing field generated by His bundle pacing.
Methods: Modelling and electrical field simulation activities were performed with the software CST (Computer Simulation Technology) from Dessault Systèms. CRT with biventricular pacing is to be achieved by an apical right ventricular electrode and an additional left ventricular electrode, which is floated into the coronary vein sinus. This conventional type of biventricular pacing leads to a reduction of the left ventricular ejection fraction. Furthermore, the non-responder rate of the CRT therapy is about one third of the CRT patients.
Results: His bundle pacing represents a physiological alternative to conventional cardiac pacing and cardiac resynchronization. An electrode implanted in the His bundle emits a stronger electrical pacing field than the electrical pacing field of conventional cardiac pacemakers. The pacing of the His bundle was performed by the Medtronic Select Secure 3830 electrode with pacing voltage amplitudes of 3 V, 2 V and 1.5 V in combination with a pacing pulse duration of 1 ms.
Conclusions: Compared to conventional cardiac pacemaker pacing, His bundle pacing is capable of bridging LBBB conduction disorders in the left ventricle. The His bundle pacing electrical field is able to spread via the physiological pathway in the right and left ventricles for CRT with a narrow QRS-complex in the surface ECG.
Patients with focal ventricular tachycardia are at risk of hemodynamic failure and if no treatment is provided the mortality rate can exceed 30%. Therefore, medical professionals must be adequately trained in the management of these conditions. To achieve the best treatment, the origin of the abnormality should be known, as well as the course of the disease. This study provides an opportunity to visualize various focal ventricular tachycardias using the Offenburg cardiac rhythm model.
Training deep neural networks using backpropagation is very memory and computationally intensive. This makes it difficult to run on-device learning or fine-tune neural networks on tiny, embedded devices such as low-power micro-controller units (MCUs). Sparse backpropagation algorithms try to reduce the computational load of on-device learning by training only a subset of the weights and biases. Existing approaches use a static number of weights to train. A poor choice of this so-called backpropagation ratio limits either the computational gain or can lead to severe accuracy losses. In this paper we present TinyProp, the first sparse backpropagation method that dynamically adapts the back-propagation ratio during on-device training for each training step. TinyProp induces a small calculation overhead to sort the elements of the gradient, which does not significantly impact the computational gains. TinyProp works particularly well on fine-tuning trained networks on MCUs, which is a typical use case for embedded applications. For typical datasets from three datasets MNIST, DCASE2020 and CIFAR10, we are 5 times faster compared to non-sparse training with an accuracy loss of on average 1%. On average, TinyProp is 2.9 times faster than existing, static sparse backpropagation algorithms and the accuracy loss is reduced on average by 6 % compared to a typical static setting of the back-propagation ratio.
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.
An Empirical Study of Explainable AI Techniques on Deep Learning Models For Time Series Tasks
(2021)
Decision explanations of machine learning black-box models are often generated by applying Explainable AI (XAI) techniques. However, many proposed XAI methods produce unverified outputs. Evaluation and verification are usually achieved with a visual interpretation by humans on individual images or text. In this preregistration, we propose an empirical study and benchmark framework to apply attribution methods for neural networks developed for images and text data on time series. We present a methodology to automatically evaluate and rank attribution techniques on time series using perturbation methods to identify reliable approaches.
Spinal cord stimulation (SCS) is the most commonly used technique of neurostimulation. It involves the stimulation of the spinal cord and is therefore used to treat chronic pain. The existing esophageal catheters are used for temperature monitoring during an electrophysiology study with ablation and transesophageal echocardiography. The aim of the study was to model the spine and new esophageal electrodes for the transesophageal electrical pacing of the spinal cord, and to integrate them in the Offenburg heart rhythm model for the static and dynamic simulation of transesophageal neurostimulation. The modeling and simulation were both performed with the electromagnetic and thermal simulation software CST (Computer Simulation Technology, Darmstadt). Two new esophageal catheters were modelled as well as a thoracic spine based on the dimensions of a human skeleton. The simulation of directed transesophageal neurostimulation is performed using the esophageal balloon catheter with an electric pacing potential of 5 V and a trapezoidal signal. A potential of 4.33 V can be measured directly at the electrode, 3.71 V in the myocardium at a depth of 2 mm, 2.68 V in the thoracic vertebra at a depth of 10 mm, 2.1 V in the thoracic vertebra at a depth of 50 mm and 2.09 V in the spinal cord at a depth of 70 mm. The relation between the voltage delivered to the electrodes and the voltage applied to the spinal cord is linear. Virtual heart rhythm and catheter models as well as the simulation of electrical pacing fields and electrical sensing fields allow the static and dynamic simulation of directed transesophageal electrical pacing of the spinal cord. The 3D simulation of the electrical sensing and pacing fields may be used to optimize transesophageal neurostimulation.
In this contribution, we propose an system setup for the detection andclassification of objects in autonomous driving applications. The recognition algo-rithm is based upon deep neural networks, operating in the 2D image domain. Theresults are combined with data of a stereo camera system to finally incorporatethe 3D object information into our mapping framework. The detection systemis locally running upon the onboard CPU of the vehicle. Several network archi-tectures are implemented and evaluated with respect to accuracy and run-timedemands for the given camera and hardware setup.
To demonstrate how deep learning can be applied to industrial applications with limited training data, deep learning methodologies are used in three different applications. In this paper, we perform unsupervised deep learning utilizing variational autoencoders and demonstrate that federated learning is a communication efficient concept for machine learning that protects data privacy. As an example, variational autoencoders are utilized to cluster and visualize data from a microelectromechanical systems foundry. Federated learning is used in a predictive maintenance scenario using the C-MAPSS dataset.
This paper describes a thorough analysis of using PPO to learn kick behaviors with simulated NAO robots in the simspark environment. The analysis includes an investigation of the influence of PPO hyperparameters, network size, training setups and performance in real games. We believe to improve the state of the art mainly in four points: first, the kicks are learned with a toed version of the NAO robot, second, we improve the reliability with respect to kickable area and avoidance of falls, third, the kick can be parameterized with desired distance and direction as input to the deep network and fourth, the approach allows to integrate the learned behavior seamlessly into soccer games. The result is a significant improvement of the general level of play.
One of the challenges for autonomous driving in general is to detect objects in the car's camera images. In the Audi Autonomous Driving Cup (AADC), among those objects are other cars, adult and child pedestrians and emergency vehicle lighting. We show that with recent deep learning networks we are able to detect these objects reliably on the limited Hardware of the model cars. Also, the same deep network is used to detect road features like mid lines, stop lines and even complete crossings. Best results are achieved using Faster R-CNN with Inception v2 showing an overall accuracy of 0.84 at 7 Hz.
The visualization of heart rhythm disturbance and atrial fibrillation therapy allow the optimization of new cardiac catheter ablations. With the simulation software CST (Computer Simulation Technology, Darmstadt) electromagnetic and thermal simulations can be carried out to analyze and optimize different heart rhythm disturbance and cardiac catheters for pulmonary vein isolation. Another form of visualization is provided by haptic, three-dimensional print models. These models can be produced using an additive manufacturing method, such as a 3D printer. The aim of the study was to produce a 3D print of the Offenburg heart rhythm model with a representation of an atrial fibrillation ablation procedure to improve the visualization of simulation of cardiac catheter ablation.
The basis of 3D printing was the Offenburg heart rhythm model and the associated simulation of cryoablation of the pulmonary vein. The thermal simulation shows the pulmonary vein isolation of the left inferior pulmonary vein with the cryoballoon catheter Arctic Front AdvanceTM from Medtronic. After running through the simulation, the thermal propagation during the procedure was shown in the form of different colors. The three-dimensional print models were constructed on the base of the described simulation in a CAD program. Four different 3D printers are available for this purpose in a rapid prototyping laboratory at the University of Applied Science Offenburg. Two different printing processes were used: 1. a binder jetting printer with polymer gypsum and 2. a multi-material printer with photopolymer. A final print model with additional representation of the esophagus and internal esophagus catheter was also prepared for printing.
With the help of the thermal simulation results and the subsequent evaluation, it was possible to make a conclusion about the propagation of the cold emanating from the catheter in the myocardium and the surrounding tissue. It could be measured that already 3 mm from the balloon surface into the myocardium the temperature drops to 25 °C. The simulation model was printed using two 3D printing methods. Both methods as well as the different printing materials offer different advantages and disadvantages. While the first model made of polymer gypsum can be produced quickly and cheaply, the second model made of photopolymer takes five times longer and was twice as expensive. On the other hand, the second model offers significantly better properties and was more durable overall. All relevant parts, especially the balloon catheter and the conduction, are realistically represented. Only the thermal propagation in the form of different colors is not shown on this model.
Three-dimensional heart rhythm models as well as virtual simulations allow a very good visualization of complex cardiac rhythm therapy and atrial fibrillation treatment methods. The printed models can be used for optimization and demonstration of cryoballoon catheter ablation in patients with atrial fibrillation.
In bimodal cochlear implant (CI) / hearing aid (HA) users a constant interaural time delay in the order of several milliseconds occurs due to differences in signal processing of the devices. For MED-EL CI systems in combination with different HA types, we have quantified the respective device delay mismatch (Zirn et al. 2015). In the current study, we investigate the effect of the device delay mismatch in simulated and actual bimodal listeners on sound localization accuracy.
To deal with the device delay mismatch in actual bimodal listeners we delayed the CI stimulation according to the measured HA processing delay and two other values. With all delay values highly significant improvements of the rms error in the localization task were observed compared to the test without the delay. The results help to narrow down the optimal patient-specific delay value.