Search

Jammed structures with finite shear modulus emerge from polycrystalline monodisperse and disordered bidisperse granular packings. To link these macroscopic mechanical properties with microstructural characteristics, we examine the vibrational behavior of two-dimensional polycrystalline monodisperse and disordered bidisperse systems using discrete element method simulations. The vibrational density of states (DOS) reflects structural disorder and soft modes, and we analyze this for both types of packings as they approach jamming and unjamming densities. Our results reveal that the low-frequency plateau in the DOS, observed in both polycrystalline and disordered packings, originates from nonaffine particle displacements, particularly those involving "rattlers", which are prominent near jamming and unjamming states. Although the jamming and unjamming process is irreversible, we find no evidence of expected history dependence in the DOS across any of the systems studied., Comment: 6 pages, 4 figures

Agitated granular media have a rich rheology: they exhibit Newtonian behavior at low shear rate and density, develop a yield stress at high density, and cross over to Bagnoldian shear thickening when sheared rapidly -- making them challenging to encompass in one theoretical framework. We measure the rheology of air-fluidized glass particles, spanning five orders of magnitude in shear rate, and show that all rheological regimes can be delineated by two dimensionless numbers. We propose a constitutive relation that captures all flow behaviors, qualitatively and quantitatively, in one unified framework.

Recent research has highlighted a key issue in speech deepfake detection: models trained on one set of deepfakes perform poorly on others. The question arises: is this due to the continuously improving quality of Text-to-Speech (TTS) models, i.e., are newer DeepFakes just 'harder' to detect? Or, is it because deepfakes generated with one model are fundamentally different to those generated using another model? We answer this question by decomposing the performance gap between in-domain and out-of-domain test data into 'hardness' and 'difference' components. Experiments performed using ASVspoof databases indicate that the hardness component is practically negligible, with the performance gap being attributed primarily to the difference component. This has direct implications for real-world deepfake detection, highlighting that merely increasing model capacity, the currently-dominant research trend, may not effectively address the generalization challenge.

For classification, the problem of class imbalance is well known and has been extensively studied. In this paper, we argue that imbalance in regression is an equally important problem which has so far been overlooked: Due to under- and over-representations in a data set's target distribution, regressors are prone to degenerate to naive models, systematically neglecting uncommon training data and over-representing targets seen often during training. We analyse this problem theoretically and use resulting insights to develop a first definition of imbalance in regression, which we show to be a generalisation of the commonly employed imbalance measure in classification. With this, we hope to turn the spotlight on the overlooked problem of imbalance in regression and to provide common ground for future research.

Voice faking, driven primarily by recent advances in text-to-speech (TTS) synthesis technology, poses significant societal challenges. Currently, the prevailing assumption is that unaltered human speech can be considered genuine, while fake speech comes from TTS synthesis. We argue that this binary distinction is oversimplified. For instance, altered playback speeds can be used for malicious purposes, like in the 'Drunken Nancy Pelosi' incident. Similarly, editing of audio clips can be done ethically, e.g., for brevity or summarization in news reporting or podcasts, but editing can also create misleading narratives. In this paper, we propose a conceptual shift away from the binary paradigm of audio being either 'fake' or 'real'. Instead, our focus is on pinpointing 'voice edits', which encompass traditional modifications like filters and cuts, as well as TTS synthesis and VC systems. We delineate 6 categories and curate a new challenge dataset rooted in the M-AILABS corpus, for which we present baseline detection systems. And most importantly, we argue that merely categorizing audio as fake or real is a dangerous over-simplification that will fail to move the field of speech technology forward.

Text-to-Speech (TTS) technology offers notable benefits, such as providing a voice for individuals with speech impairments, but it also facilitates the creation of audio deepfakes and spoofing attacks. AI-based detection methods can help mitigate these risks; however, the performance of such models is inherently dependent on the quality and diversity of their training data. Presently, the available datasets are heavily skewed towards English and Chinese audio, which limits the global applicability of these anti-spoofing systems. To address this limitation, this paper presents the Multi-Language Audio Anti-Spoof Dataset (MLAAD), created using 82 TTS models, comprising 33 different architectures, to generate 378.0 hours of synthetic voice in 38 different languages. We train and evaluate three state-of-the-art deepfake detection models with MLAAD and observe that it demonstrates superior performance over comparable datasets like InTheWild and Fake- OrReal when used as a training resource. Moreover, compared to the renowned ASVspoof 2019 dataset, MLAAD proves to be a complementary resource. In tests across eight datasets, MLAAD and ASVspoof 2019 alternately outperformed each other, each excelling on four datasets. By publishing MLAAD and making a trained model accessible via an interactive webserver, we aim to democratize anti-spoofing technology, making it accessible beyond the realm of specialists, and contributing to global efforts against audio spoofing and deepfakes., Comment: IJCNN 2024

Neural networks build the foundation of several intelligent systems, which, however, are known to be easily fooled by adversarial examples. Recent advances made these attacks possible even in air-gapped scenarios, where the autonomous system observes its surroundings by, e.g., a camera. We extend these ideas in our research and evaluate the robustness of multi-camera setups against such physical adversarial examples. This scenario becomes ever more important with the rise in popularity of autonomous vehicles, which fuse the information of several cameras for their driving decision. While we find that multi-camera setups provide some robustness towards past attack methods, we see that this advantage reduces when optimizing on multiple perspectives at once. We propose a novel attack method that we call Transcender-MC, where we incorporate online 3D renderings and perspective projections in the training process. Moreover, we motivate that certain data augmentation techniques can facilitate the generation of successful adversarial examples even further. Transcender-MC is 11% more effective in successfully attacking multi-camera setups than state-of-the-art methods. Our findings offer valuable insights regarding the resilience of object detection in a setup with multiple cameras and motivate the need of developing adequate defense mechanisms against them.

Machine-learning (ML) shortcuts or spurious correlations are artifacts in datasets that lead to very good training and test performance but severely limit the model's generalization capability. Such shortcuts are insidious because they go unnoticed due to good in-domain test performance. In this paper, we explore the influence of different shortcuts and show that even simple shortcuts are difficult to detect by explainable AI methods. We then exploit this fact and design an approach to defend online databases against crawlers: providers such as dating platforms, clothing manufacturers, or used car dealers have to deal with a professionalized crawling industry that grabs and resells data points on a large scale. We show that a deterrent can be created by deliberately adding ML shortcuts. Such augmented datasets are then unusable for ML use cases, which deters crawlers and the unauthorized use of data from the internet. Using real-world data from three use cases, we show that the proposed approach renders such collected data unusable, while the shortcut is at the same time difficult to notice in human perception. Thus, our proposed approach can serve as a proactive protection against illegitimate data crawling., Comment: Published at BMVC 2023

Fluidized granular media have a rich rheology: measuring shear stress $\sigma$ as a function of shear rate $\dot\gamma$, they exhibit Newtonian behavior $\sigma\sim\dot\gamma$ for low densities and shear rates, develop a yield stress for intermediate shear rates and densities approaching the granular glass transition, and finally, cross over to shear-thickening Bagnold scaling, $\sigma\sim\dot\gamma^2$. This wealth of flow-behaviors makes fluidized beds a fascinating material, but also one that is challenging to encompass into a global theory, despite its relevance for optimizing industrial processes and predicting natural hazards. We provide careful measurements spanning eight orders of magnitude in shear rate, and show that all these rheological regimes can be described qualitatively and quantitatively using the granular integration through transient formalism, a theory for glassy dynamics under shear adapted to granular fluids.

Current anti-spoofing and audio deepfake detection systems use either magnitude spectrogram-based features (such as CQT or Melspectrograms) or raw audio processed through convolution or sinc-layers. Both methods have drawbacks: magnitude spectrograms discard phase information, which affects audio naturalness, and raw-feature-based models cannot use traditional explainable AI methods. This paper proposes a new approach that combines the benefits of both methods by using complex-valued neural networks to process the complex-valued, CQT frequency-domain representation of the input audio. This method retains phase information and allows for explainable AI methods. Results show that this approach outperforms previous methods on the "In-the-Wild" anti-spoofing dataset and enables interpretation of the results through explainable AI. Ablation studies confirm that the model has learned to use phase information to detect voice spoofing., Comment: Interspeech 2023

User experience (UX) is a part of human-computer interaction (HCI) research and focuses on increasing intuitiveness, transparency, simplicity, and trust for the system users. Most UX research for machine learning (ML) or natural language processing (NLP) focuses on a data-driven methodology. It engages domain users mainly for usability evaluation. Moreover, more typical UX methods tailor the systems towards user usability, unlike learning about the user needs first. This paper proposes a new methodology for integrating generative UX research into developing domain NLP applications. Generative UX research employs domain users at the initial stages of prototype development, i.e., ideation and concept evaluation, and the last stage for evaluating system usefulness and user utility. The methodology emerged from and is evaluated on a case study about the full-cycle prototype development of a domain-specific semantic search for daily operations in the process industry. A key finding of our case study is that involving domain experts increases their interest and trust in the final NLP application. The combined UX+NLP research of the proposed method efficiently considers data- and user-driven opportunities and constraints, which can be crucial for developing NLP applications.

An efficient approach for the construction of separable approximations of optimal value functions from interconnected optimal control problems is presented. The approach is based on assuming decaying sensitivities between subsystems, enabling a curse-of-dimensionality free approximation, for instance by deep neural networks.

Pulmonary Embolisms (PE) represent a leading cause of cardiovascular death. While medical imaging, through computed tomographic pulmonary angiography (CTPA), represents the gold standard for PE diagnosis, it is still susceptible to misdiagnosis or significant diagnosis delays, which may be fatal for critical cases. Despite the recently demonstrated power of deep learning to bring a significant boost in performance in a wide range of medical imaging tasks, there are still very few published researches on automatic pulmonary embolism detection. Herein we introduce a deep learning based approach, which efficiently combines computer vision and deep neural networks for pulmonary embolism detection in CTPA. Our method features novel improvements along three orthogonal axes: 1) automatic detection of anatomical structures; 2) anatomical aware pretraining, and 3) a dual-hop deep neural net for PE detection. We obtain state-of-the-art results on the publicly available multicenter large-scale RSNA dataset., Comment: Accepted to Computers in Biology and Medicine journal

For real-world applications of machine learning (ML), it is essential that models make predictions based on well-generalizing features rather than spurious correlations in the data. The identification of such spurious correlations, also known as shortcuts, is a challenging problem and has so far been scarcely addressed. In this work, we present a novel approach to detect shortcuts in image and audio datasets by leveraging variational autoencoders (VAEs). The disentanglement of features in the latent space of VAEs allows us to discover feature-target correlations in datasets and semi-automatically evaluate them for ML shortcuts. We demonstrate the applicability of our method on several real-world datasets and identify shortcuts that have not been discovered before., Comment: Accepted at the ICML 2023 Workshop on Spurious Correlations, Invariance and Stability

We study the local structural changes along the jamming transitions in asymmetric bidisperse granu\-lar packings. The local structure of the packing is assessed by the contact orientational order, $\tilde{Q}_{\ell}$, that quantifies the contribution of each contact configuration (Large-Large, Small-Small, Large-Small, Small-Large) in the jammed structure. The partial values of $\tilde{Q}_{\ell}$ are calculated with respect to known ordered lattices that are fixed by the size ratio, $\delta$, of the particles. We find that the packing undergoes a structural transition at $\phi_J$, manifested by a sudden jump in the partial $\tilde{Q}_{\ell}$. Each contact configuration contributes to the jammed structure in a different way, changing with $\delta$ and concentration of small particles, $X_{\mathrm{S}}$. The results show not only that the packing undergoes a structural change upon jamming, but also that bidisperse packings exhibit local HCP and FCC structures also found in monodisperse packings. This suggests that the jammed structure of bidisperse systems is inherently endowed with local structural order. These results are relevant in understanding how the arrangement of particles determines the strength of bidisperse granular packings., Comment: 13 pages, 14 figures. Submitted to Granular Matter

Neural networks follow a gradient-based learning scheme, adapting their mapping parameters by back-propagating the output loss. Samples unlike the ones seen during training cause a different gradient distribution. Based on this intuition, we design a novel semi-supervised anomaly detection method called R2-AD2. By analysing the temporal distribution of the gradient over multiple training steps, we reliably detect point anomalies in strict semi-supervised settings. Instead of domain dependent features, we input the raw gradient caused by the sample under test to an end-to-end recurrent neural network architecture. R2-AD2 works in a purely data-driven way, thus is readily applicable in a variety of important use cases of anomaly detection., Comment: Accepted at ECML-PKDD 2022

This paper introduces Finite Elements with Switch Detection (FESD), a numerical discretization method for nonsmooth differential equations. We consider the Filippov convexification of these systems and a transformation into dynamic complementarity systems introduced by [Stewart, 1990]. FESD is based on solving nonlinear complementarity problems and can automatically detect nonsmooth events in time. If standard time-stepping Runge-Kutta (RK) methods are naively applied to a nonsmooth ODE, the accuracy is at best of order one. In FESD, we let the integrator step size be a degree of freedom. Additional complementarity conditions, which we call cross complementarities, enable exact switch detection, hence FESD can recover the high order accuracy that the RK methods enjoy for smooth ODE. Additional conditions called step equilibration allow the step size to change only when switches occur and thus avoid spurious degrees of freedom. Convergence results for the FESD method are derived, local uniqueness of the solution and convergence of numerical sensitivities are proven. The efficacy of FESD is demonstrated in several simulation and optimal control examples. In an optimal control problem benchmark with FESD, we achieve up to five orders of magnitude more accurate solutions than a standard time-stepping approach for the same computational time., Comment: Under review in Numerische Mathematik

The macroscopic response of granular solids is determined by the microscopic fabric of force chains, which, in turn, is intimately linked to the history of the solid. To query the influence of gravity on powder flow-behavior, a granular material is subjected to compression by a piston in a closed container, on-ground and in microgravity. Results show that piston-probing densifies the packing, eventually leading to jamming of the material compressed by the piston, regardless of the gravitational environment. The onset of jamming is found to appear at lower packing fraction in microgravity ($\varphi^{\textrm{$\mu$-g}}_J = 0.567 \pm 0.014$) than on-ground ($\varphi^{\text{gnd}}_J = 0.579 \pm 0.014 $). We interpret these findings as the manifestation of a granular fabric altered by the gravitational force field: in absence of a secondary load (due to gravitational acceleration) to stimulate reorganization in a different direction to the major compression stress, the particles' configuration becomes stable at lower density, as the particles have no external drive to promote reorganization into a denser packing. This is coupled with a change in interparticular force balance which takes place under low gravity, as cohesive interactions become predominant. We propose a combination of microscopic and continuum arguments to rationalize our results., Comment: npj Microgravity (2022)

Automatic speech recognition (ASR) is improving ever more at mimicking human speech processing. The functioning of ASR, however, remains to a large extent obfuscated by the complex structure of the deep neural networks (DNNs) they are based on. In this paper, we show how so-called attribution methods, that we import from image recognition and suitably adapt to handle audio data, can help to clarify the working of ASR. Taking DeepSpeech, an end-to-end model for ASR, as a case study, we show how these techniques help to visualize which features of the input are the most influential in determining the output. We focus on three visualization techniques: Layer-wise Relevance Propagation (LRP), Saliency Maps, and Shapley Additive Explanations (SHAP). We compare these methods and discuss potential further applications, such as in the detection of adversarial examples., Comment: Proc. 2021 ISCA Symposium on Security and Privacy in Speech Communication

We investigate, using a recently developed model of liquid state theory describing the rheology of dense granular flows, how a yield stress appears in granular matter at the yielding transition. Our model allows us to predict an analytical equation of the corresponding dynamical yield surface, which is compared to usual models of solid fracture. In particular, this yield surface interpolates between the typical failure behaviors of soft and hard materials. This work also underlines the central role played by the effective friction coefficient at the yielding transition., Comment: 13 pages, 12 figures

An important problem in deep learning is the privacy and security of neural networks (NNs). Both aspects have long been considered separately. To date, it is still poorly understood how privacy enhancing training affects the robustness of NNs. This paper experimentally evaluates the impact of training with Differential Privacy (DP), a standard method for privacy preservation, on model vulnerability against a broad range of adversarial attacks. The results suggest that private models are less robust than their non-private counterparts, and that adversarial examples transfer better among DP models than between non-private and private ones. Furthermore, detailed analyses of DP and non-DP models suggest significant differences between their gradients. Additionally, this work is the first to observe that an unfavorable choice of parameters in DP training can lead to gradient masking, and, thereby, results in a wrong sense of security.

The Granular Integration Through Transients (GITT) formalism gives a theoretical description of the rheology of moderately dense granular flows and suspensions. In this work, we extend the GITT equations beyond the case of simple shear flows studied before. Applying this to the particular example of extensional flows, we show that the predicted behavior is somewhat different from that of the more frequently studied simple shear case, as illustrated by the possibility of non monotonous evolution of the effective friction coefficient $\mu$ with the inertial number $\mathcal{I}$. By the reduction of the GITT equations to simple toy-models, we provide a generalization of the $\mu(\mathcal{I})$-law true for any type of flow deformation. Our analysis also includes a study of the Trouton ratio, which is shown to behave quite similarly to that of dense colloidal suspensions., Comment: 17 pages, 11 figures

We present 3D DEM simulations of bidisperse granular packings to investigate their jamming densities, $\phi_J$, and dimensionless bulk moduli, $K$, as a function of the size ratio, $\delta$, and the concentration of small particles, $X_{\mathrm S}$. We determine the partial and total bulk moduli for each packing and report the jamming transition diagram, i.e., the density or volume fraction marking both the first and second transitions of the system. At a large enough size difference, e.g., $\delta \le 0.22$, $X^{*}_{\mathrm S}$ divides the diagram with most small particles either non-jammed or jammed jointly with large ones. We find that the bulk modulus $K$ jumps at $X^{*}_{\mathrm S}(\delta = 0.15) \approx 0.21$, at the maximum jamming density, where both particle species mix most efficiently, while for $X_{\mathrm S} < X^{*}_{\mathrm S}$ $K$ is decoupled in two scenarios as a result of the first and second jamming transition. Along the second transition, $K$ rises relative to the values found at the first transition, however, is still small compared to $K$ at $X^{*}_{\mathrm S}$. While the first transition is sharp, the second is smooth, carried by small-large interactions, while the small-small contacts display a transition. This demonstrates that for low enough $\delta$ and $X_{\mathrm S}$, the jamming of small particles indeed impacts the internal resistance of the system. Our new results will allow tuning the bulk modulus $K$ or other properties, such as the wave speed, by choosing specific sizes and concentrations based on a better understanding of whether small particles contribute to the jammed structure or not, and how the micromechanical structure behaves at either transition., Comment: 15 pages, 18 figures

Anomaly detection is a challenging task for machine learning algorithms due to the inherent class imbalance. It is costly and time-demanding to manually analyse the observed data, thus usually only few known anomalies if any are available. Inspired by generative models and the analysis of the hidden activations of neural networks, we introduce a novel unsupervised anomaly detection method called DA3D. Here, we use adversarial autoencoders to generate anomalous counterexamples based on the normal data only. These artificial anomalies used during training allow the detection of real, yet unseen anomalies. With our novel generative approach, we transform the unsupervised task of anomaly detection to a supervised one, which is more tractable by machine learning and especially deep learning methods. DA3D surpasses the performance of state-of-the-art anomaly detection methods in a purely data-driven way, where no domain knowledge is required., Comment: Accepted at IJCNN 2022

Anomaly detection has been considered under several extents of prior knowledge. Unsupervised methods do not require any labelled data, whereas semi-supervised methods leverage some known anomalies. Inspired by mixture-of-experts models and the analysis of the hidden activations of neural networks, we introduce a novel data-driven anomaly detection method called ARGUE. Our method is not only applicable to unsupervised and semi-supervised environments, but also profits from prior knowledge of self-supervised settings. We designed ARGUE as a combination of dedicated expert networks, which specialise on parts of the input data. For its final decision, ARGUE fuses the distributed knowledge across the expert systems using a gated mixture-of-experts architecture. Our evaluation motivates that prior knowledge about the normal data distribution may be as valuable as known anomalies., Comment: Accepted at IJCNN 2022

We adapt statistical models of the physics of complex fluids to study the rheology of granular liquids. This allows us to provide laws of granular rheology based on first principles, which compare well with previously established phenomenological laws. In particular, the very successful law of mu(I) rheology can be understood within our model as the lowest order non trivial Pade approximant of the macroscopic laws of rheology. Our model's ability to describe granular physics outside of the Bagnold scaling regime allows for a natural extension to the rheology of granular suspensions., Comment: 13 pages, 5 figures