arXiv subjects

Alexander Jung

Publications and source records attributed to Alexander Jung.

At least 19 recordsLinked to original sources

arXiv2026

Byzantine-Robust Federated Fire Detection with a Rotating Coordinator

We study the application of federated learning (FL) to indoor fire detection. Such fire-detection systems use edge cameras that record sensitive footage which cannot easily be collected at a central server. Existing federated solutions leave three practical obstacles unaddressed: limited uplink bandwidth, Byzantine (malicious or faulty) clients, and unconditional trust in a single, permanently fixed aggregation server. Our main contributions address all three. In particular, we provide (i) a curated indoor fire-detection dataset assembled from eight public sources; (ii) an edge-deployable detector whose model updates are compressed up to 10 time with only a small loss in balanced accuracy; and (iii) a semi-decentralized Byzantine-robust FL method that combines history-aware aggregation with a rotating coordinator, evicting stealthy attacks that per-round filters miss while removing the fixed-server single point of failure. On the held-out test set the rotating-coordinator method matches its fixed-server counterpart in accuracy and detection speed, and a physically distributed six-node cloud deployment confirms feasibility.

cs.LG↗

arXiv2026

$\alpha$-TCAV: A Unified Framework for Testing with Concept Activation Vectors

Concept Activation Vectors (CAVs) are a fundamental tool for concept-based explainability in deep learning, yet their practical utility is limited by statistical instability. We analyze the stochastic nature of CAVs and the Testing with CAVs (TCAV) method, deriving the distributions of major CAV classes including PatternCAV, FastCAV, and ridge regression-based CAVs. We then identify a fundamental flaw in the standard TCAV score: its reliance on a discontinuous indicator function induces non-decaying variance in critical regimes. To address this, we introduce $\alpha$-TCAV, a generalized framework that replaces the indicator with a parameterized smooth function, yielding a unified probabilistic formulation that subsumes both TCAV and Multi-TCAV. We characterize the induced distributions of sensitivity scores and different TCAV variants, showing that established state-of-the-art choices lack theoretical justification. We provide principled guidance on tuning the parameter in $\alpha$-TCAV -- either to imitate Multi-TCAV at substantially lower computational cost, or to obtain a calibrated Bayes-optimal probabilistic measure of a concept's influence. Finally, our analysis yields practical recommendations that challenge established routines: most notably, allocating the full sampling budget to a single CAV rather than splitting it across several.

stat.ML↗

arXiv2026

Wayfinder: Automated Operating System Specialization

Specializing an OS to optimize the performance of a particular application is typically a manual process that requires great expertise. Specialization through configuration lends itself well to automation; however, it is challenging due to the sheer size of the configuration space of modern OSes, the difficulty to quantify that space, the long time it takes to evaluate a configuration, and the large number of invalid configurations. Hence, existing attempts at specializing OSes automatically are limited to switching features on and off to minimize memory consumption or attack surface, and cannot target metrics such as performance. We present Wayfinder, a framework specializing the configuration of OSes completely automatically and without expert knowledge. It can specialize all aspects of an OS configuration (compile-/boot-/run-time) towards any quantifiable performance, resource consumption, or security metric, for an application processing a given workload on a given hardware setup. Wayfinder consists of an automated OS benchmarking platform, and a neural network-based search algorithm driving the specialization process. This is achieved by learning on the fly which configuration parameters and values impact performance the most, and which ones lead to runtime failures. Optionally, a model pre-trained on one application can be reused to accelerate the specialization of related applications. We evaluate Wayfinder on two OSes, four applications, and two target metrics: Wayfinder fully automatically identifies specialized configurations with up to 24% application performance improvement and 8.5% memory usage reduction compared to default configurations. We highlight the benefits of our neural network, reaching good solutions faster than competing approaches (random and Bayesian), and successfully transferring knowledge between related applications.

cs.OS↗

arXiv2026

Interpretable Multiple Myeloma Prognosis with Observational Medical Outcomes Partnership Data

Machine learning (ML) promises better clinical decision-making, yet opaque model behavior limits the adoption in healthcare. We propose two novel regularization techniques for ensuring the interpretability of ML models trained on real-world data. In particular, we consider the prediction of five-year survival for multiple myeloma patients using clinical data from Helsinki University Hospital. To ensure the interpretability of the trained models, we use two alternative constructions for a penalty term used for regularization. The first one penalizes deviations from the predictions obtained from an interpretable logistic regression method with two manually chosen features. The second construction requires consistency of model predictions with the revised international staging system (R-ISS). We verify the usefulness of the proposed regularization techniques in numerical experiments using data from 812 patients. They achieve an accuracy up to 0.721 on a test set and SHAP values show that the models rely on the selected important features.

cs.LG↗

arXiv2025

Federated k-Means over Networks

We study federated clustering, where interconnected devices collaboratively cluster the data points of private local datasets. Focusing on hard clustering via the k-means principle, we formulate federated k-means as an instance of generalized total variation minimization (GTVMin). This leads to a federated k-means algorithm in which each device updates its local cluster centroids by solving a regularized k-means problem with a regularizer that enforces consistency between neighbouring devices. The resulting algorithm is privacy-friendly, as only aggregated information is exchanged.

cs.LG↗

arXiv2025

Membrane phononic crystals for high-Qm mechanical defect modes in piezoelectric aluminum nitride

Nanomechanical resonators with exceptionally low dissipation are advancing mechanics-based sensors and quantum technologies. The key for these advances is the engineering of localized phononic modes that are well-isolated from the environment, i.e., that exhibit a high mechanical quality factor, Qm. Membrane phononic crystals fabricated from strained thin films can realize high-Qm single or multiple localized phononic defect modes at MHz frequencies. These defect modes can be efficiently interfaced with out-of-plane light or coupled to a microwave quantum circuit, enabling readout and control of their motion. When membrane phononic crystals are fabricated from a crystalline film, they could offer built-in functionality. We demonstrate a membrane phononic crystal realized in a strained 90 nm-thin film of aluminum nitride (AlN), which is a crystalline piezoelectric material. We engineer a high-Qm localized phononic defect mode at 1.8 MHz with a Qxf-product of 1.5x10^13 Hz at room temperature. In future devices, the built-in piezoelectricity of AlN can be utilized for direct coupling to qubits or in-situ tuning of mechanical mode frequencies, defect mode couplings, or acoustic bandgaps, which can be used as building blocks of tunable phononic circuits or low-noise sensors.

physics.app-ph↗

arXiv2025

Enforcing Fundamental Relations via Adversarial Attacks on Input Parameter Correlations

Correlations between input parameters play a crucial role in many scientific classification tasks, since these are often related to fundamental laws of nature. For example, in high energy physics, one of the common deep learning use-cases is the classification of signal and background processes in particle collisions. In many such cases, the fundamental principles of the correlations between observables are often better understood than the actual distributions of the observables themselves. In this work, we present a new adversarial attack algorithm called Random Distribution Shuffle Attack (RDSA), emphasizing the correlations between observables in the network rather than individual feature characteristics. Correct application of the proposed novel attack can result in a significant improvement in classification performance - particularly in the context of data augmentation - when using the generated adversaries within adversarial training. Given that correlations between input features are also crucial in many other disciplines. We demonstrate the RDSA effectiveness on six classification tasks, including two particle collision challenges (using CERN Open Data), hand-written digit recognition (MNIST784), human activity recognition (HAR), weather forecasting (Rain in Australia), and ICU patient mortality (MIMIC-IV), demonstrating a general use case beyond fundamental physics for this new type of adversarial attack algorithms.

cs.LG↗

arXiv2024

Engineering Trustworthy AI: A Developer Guide for Empirical Risk Minimization

AI systems increasingly shape critical decisions across personal and societal domains. While empirical risk minimization (ERM) drives much of the AI success, it typically prioritizes accuracy over trustworthiness, often resulting in biases, opacity, and other adverse effects. This paper discusses how key requirements for trustworthy AI can be translated into design choices for the components of ERM. We hope to provide actionable guidance for building AI systems that meet emerging standards for trustworthiness of AI.

cs.AI↗

arXiv2024

Thickness dependence of the mechanical properties of piezoelectric high-$Q_m$ nanomechanical resonators made from aluminium nitride

Nanomechanical resonators with high quality factors (\Qm{}) enable mechanics-based quantum technologies, in particular quantum sensing and quantum transduction. High-\Qm{} nanomechanical resonators in the kHz to MHz frequency range can be realized in tensile-strained thin films that allow the use of dissipation dilution techniques to drastically increase \Qm{}. In our work, we study the material properties of tensile-strained piezoelectric films made from aluminium nitride (AlN). We characterize crystalline AlN films with a thickness ranging from \SI{45}{\nano\meter} to \SI{295}{\nano\meter}, which are directly grown on Si(111) by metal-organic vapour-phase epitaxy. We report on the crystal quality and surface roughness, the piezoelectric response, and the residual and released stress of the AlN thin films. Importantly, we determine the intrinsic quality factor of the films at room temperature in high vacuum. We fabricate and characterize AlN nanomechanical resonators that exploit dissipation dilution to enhance the intrinsic quality factor by utilizing the tensile strain in the film. We find that AlN nanomechanical resonators below \SI{200}{\nano\meter} thickness exhibit the highest \Qf{}-product, on the order of $10^{12}$\,Hz. We discuss possible strategies to optimize the material growth that should lead to devices that reach even higher \Qf{}-products. This will pave the way for future advancements of optoelectromechanical quantum devices made from tensile-strained piezoelectric AlN.

cond-mat.mes-hall↗

arXiv2024

Your Data, My Model: Learning Who Really Helps in Federated Learning

Many important machine learning applications involve networks of devices-such as wearables or smartphones-that generate local data and train personalized models. A key challenge is determining which peers are most beneficial for collaboration. We propose a simple and privacy-preserving method to select relevant collaborators by evaluating how much a model improves after a single gradient step using another devices data-without sharing raw data. This method naturally extends to non-parametric models by replacing the gradient step with a non-parametric generalization. Our approach enables model-agnostic, data-driven peer selection for personalized federated learning (PersFL).

cs.LG↗

arXiv2024

Inertia emulation contribution of Frades 2 variable speed pump-turbine to power network stability

This paper is addressing the quantification and the comparison of pumped storage power plants, PSPP, contribution to synchronous inertia and synthetic inertia when fixed speed and variable speed motor-generators technologies are considered, respectively. Therefore, a grid stability study was conducted by means of 1D SIMSEN simulation for the 2 x 395 MW PSPP Frades 2 in Portugal with both fixed speed and variable speed technologies in case of operation connected to an infinite power network or to an islanded 4.4 GW synchronous power network.

eess.SY↗

arXiv2024

Nanomechanical crystalline AlN resonators with high quality factors for quantum optoelectromechanics

High-\Qm{} mechanical resonators are crucial for applications where low noise and long coherence time are required, as mirror suspensions, quantum cavity optomechanical devices, or nanomechanical sensors. Tensile strain in the material enables the use of dissipation dilution and strain engineering techniques, which increase the mechanical quality factor. These techniques have been employed for high-\Qm{} mechanical resonators made from amorphous materials and, recently, from crystalline materials such as InGaP, SiC, and Si. A strained crystalline film exhibiting substantial piezoelectricity expands the capability of high-\Qm{} nanomechanical resonators to directly utilize electronic degrees of freedom. In this work we realize nanomechanical resonators with \Qm{} up to $2.9\times 10^{7}$ made from tensile-strained \SI{290}{\nano\meter}-thick AlN, which is an epitaxially-grown crystalline material offering strong piezoelectricity. We demonstrate nanomechanical resonators that exploit dissipation dilution and strain engineering to reach a \Qf-product approaching $10^{13}$\,\SI{}{\hertz} at room temperature. We realize a novel resonator geometry, triangline, whose shape follows the Al-N bonds and offers a central pad that we pattern with a photonic crystal. This allows us to reach an optical reflectivity above 80\% for efficient coupling to out-of-plane light. The presented results pave the way for quantum optoelectromechanical devices at room temperature based on tensile-strained AlN.

cond-mat.mes-hall↗

arXiv2023

Moreau Envelope ADMM for Decentralized Weakly Convex Optimization

This paper proposes a proximal variant of the alternating direction method of multipliers (ADMM) for distributed optimization. Although the current versions of ADMM algorithm provide promising numerical results in producing solutions that are close to optimal for many convex and non-convex optimization problems, it remains unclear if they can converge to a stationary point for weakly convex and locally non-smooth functions. Through our analysis using the Moreau envelope function, we demonstrate that MADM can indeed converge to a stationary point under mild conditions. Our analysis also includes computing the bounds on the amount of change in the dual variable update step by relating the gradient of the Moreau envelope function to the proximal function. Furthermore, the results of our numerical experiments indicate that our method is faster and more robust than widely-used approaches.

math.OC↗

arXiv2023

Surface acoustic wave resonators on thin film piezoelectric substrates in the quantum regime

Lithium niobate (LNO) is a well established material for surface acoustic wave (SAW) devices including resonators, delay lines and filters. Recently, multi-layer substrates based on LNO thin films have become commercially available. Here, we present a systematic low-temperature study of the performance of SAW devices fabricated on LNO-on-insulator and LNO-on-Silicon substrates and compare them to bulk LNO devices. Our study aims at assessing the performance of these substrates for quantum acoustics, i.e. the integration with superconducting circuits operating in the quantum regime. To this end, we design SAW resonators with a target frequency of 5 GHz and perform experiments at millikelvin temperatures and microwave power levels corresponding to single photons or phonons. The devices are investigated regarding their internal quality factors as a function of the excitation power and temperature, which allows us to characterize and quantify losses and identify the dominating loss mechanism. For the measured devices, fitting the experimental data shows that the quality factors are limited by the coupling of the resonator to a bath of two-level-systems. Our results suggest that SAW devices on thin film LNO on silicon have comparable performance to devices on bulk LNO and are viable for use in SAW-based quantum acoustic devices.

physics.app-ph↗

arXiv2022

Machine Learning: The Basics

Machine learning (ML) has become a commodity in our every-day lives. We routinely ask ML empowered smartphones to suggest lovely food places or to guide us through a strange place. ML methods have also become standard tools in many fields of science and engineering. A plethora of ML applications transform human lives at unprecedented pace and scale. This book portrays ML as the combination of three basic components: data, model and loss. ML methods combine these three components within computationally efficient implementations of the basic scientific principle "trial and error". This principle consists of the continuous adaptation of a hypothesis about a phenomenon that generates data. ML methods use a hypothesis to compute predictions for future events. We believe that thinking about ML as combinations of three components given by data, model, and loss helps to navigate the steadily growing offer for ready-to-use ML methods. Our three-component picture of ML allows a unified treatment of a wide range of concepts and techniques which seem quite unrelated at first sight. The regularization effect of early stopping in iterative methods is due to the shrinking of the effective hypothesis space. Privacy-preserving ML is obtained by particular choices for the features of data points. Explainable ML methods are characterized by particular choices for the hypothesis space. To make good use of ML tools it is instrumental to understand its underlying principles at different levels of detail. On a lower level, this tutorial helps ML engineers to choose suitable methods for the application at hand. The book also offers a higher-level view on the implementation of ML methods which is typically required to manage a team of ML engineers and data scientists.

cs.LG↗

arXiv2022

FlexOS: Towards Flexible OS Isolation

At design time, modern operating systems are locked in a specific safety and isolation strategy that mixes one or more hardware/software protection mechanisms (e.g. user/kernel separation); revisiting these choices after deployment requires a major refactoring effort. This rigid approach shows its limits given the wide variety of modern applications' safety/performance requirements, when new hardware isolation mechanisms are rolled out, or when existing ones break. We present FlexOS, a novel OS allowing users to easily specialize the safety and isolation strategy of an OS at compilation/deployment time instead of design time. This modular LibOS is composed of fine-grained components that can be isolated via a range of hardware protection mechanisms with various data sharing strategies and additional software hardening. The OS ships with an exploration technique helping the user navigate the vast safety/performance design space it unlocks. We implement a prototype of the system and demonstrate, for several applications (Redis/Nginx/SQLite), FlexOS' vast configuration space as well as the efficiency of the exploration technique: we evaluate 80 FlexOS configurations for Redis and show how that space can be probabilistically subset to the 5 safest ones under a given performance budget. We also show that, under equivalent configurations, FlexOS performs similarly or better than several baselines/competitors.

cs.OS↗

arXiv2021

Rethinking Drone-Based Search and Rescue with Aerial Person Detection

The visual inspection of aerial drone footage is an integral part of land search and rescue (SAR) operations today. Since this inspection is a slow, tedious and error-prone job for humans, we propose a novel deep learning algorithm to automate this aerial person detection (APD) task. We experiment with model architecture selection, online data augmentation, transfer learning, image tiling and several other techniques to improve the test performance of our method. We present the novel Aerial Inspection RetinaNet (AIR) algorithm as the combination of these contributions. The AIR detector demonstrates state-of-the-art performance on a commonly used SAR test data set in terms of both precision (~21 percentage point increase) and speed. In addition, we provide a new formal definition for the APD problem in SAR missions. That is, we propose a novel evaluation scheme that ranks detectors in terms of real-world SAR localization requirements. Finally, we propose a novel postprocessing method for robust, approximate object localization: the merging of overlapping bounding boxes (MOB) algorithm. This final processing stage used in the AIR detector significantly improves its performance and usability in the face of real-world aerial SAR missions.

cs.CV↗

arXiv2021

Clustered Federated Learning via Generalized Total Variation Minimization

We study optimization methods to train local (or personalized) models for decentralized collections of local datasets with an intrinsic network structure. This network structure arises from domain-specific notions of similarity between local datasets. Examples for such notions include spatio-temporal proximity, statistical dependencies or functional relations. Our main conceptual contribution is to formulate federated learning as generalized total variation (GTV) minimization. This formulation unifies and considerably extends existing federated learning methods. It is highly flexible and can be combined with a broad range of parametric models, including generalized linear models or deep neural networks. Our main algorithmic contribution is a fully decentralized federated learning algorithm. This algorithm is obtained by applying an established primal-dual method to solve GTV minimization. It can be implemented as message passing and is robust against inexact computations that arise from limited computational resources including processing time or bandwidth. Our main analytic contribution is an upper bound on the deviation between the local model parameters learnt by our algorithm and an oracle-based clustered federated learning method. This upper bound reveals conditions on the local models and the network structure of local datasets such that GTV minimization is able to pool (nearly) homogeneous local datasets.

cs.LG↗