arXiv subjects

Marius Köppel

Publications and source records attributed to Marius Köppel.

16 recordsLinked to original sources

arXiv2026

Letter Of Intent for a future $μ^+ \to \mathrm{e}^+ γ$ experiment at the High Intensity Muon Beam facility at PSI

Searches for charged lepton flavor violation in the muon sector stand out among the most sensitive and clean probes for physics beyond the Standard Model. Currently, $μ^+ \to \mathrm{e}^+ γ$ experiments provide the best constraints in this field for a wide range of models while, in the coming years, new experiments investigating the processes of $μ^+ \to \mathrm{e}^+ \mathrm{e}^+ \mathrm{e}^-$ and $μ\to \mathrm{e}$ conversion in the nuclear field are anticipated to reach comparable or higher sensitivities. The High-Intensity Muon Beam (HIMB) facility at PSI, which is expected to deliver muon beam intensities up to two orders of magnitude higher than the existing beam lines, offers a unique opportunity to significantly enhance the sensitivity of $μ^+ \to \mathrm{e}^+ γ$ searches. The discovery potential could be substantially boosted and a sensitivity comparable to that of all the other projects could be reestablished, which is essential for discriminating among competing new-physics scenarios should an observation occur in any of the channels. In this document, we express our interest in developing a $μ^+ \to \mathrm{e}^+ γ$ experimental program at HIMB, with the goal of improving, within the next decade, the sensitivity of the $μ^+ \to \mathrm{e}^+ γ$ search by more than one order of magnitude relative to the expected final result of the current leading experiment, MEG II. This effort would ensure that PSI retains its leadership in this field.

hep-ex↗

arXiv2025

New Frontiers in Muon-Spin Spectroscopy Using Si-Pixel Detectors

The study of novel quantum materials relies on muon-spin rotation, relaxation, or resonance (\mSR) measurements. Yet, a fundamental limitation persists: many of these materials can only be synthesized in extremely small quantities, often at sub-millimeter scales. While \mSR ~offers unique insights into electronic and magnetic properties, existing spectrometers lack a sub-millimeter spatial resolution and the possibility of triggerless pump-probe data acquisition, which would enable more advanced measurements. The General Purpose Surface-muon instrument (GPS) at the Paul Scherrer Institute (PSI) is currently limited to a muon stopping rate of \SI{40}{\kilo\hertz} to \SI{120}{\kilo\hertz}, a constraint that will become more pressing with the upcoming High-Intensity Muon Beam (HIMB) project. To overcome these challenges, we demonstrate the feasibility of employing ultra-thin monolithic Si-pixel detectors to reconstruct the stopping position of muons within the sample, thereby significantly enhancing the capability of measuring at higher muon rate. Additionally, we explore the first steps toward a triggerless pump-probe \mSR ~measurement scheme. Unlike conventional pump-probe techniques that require external triggers, a triggerless readout system can continuously integrate stimuli pulses into the data stream, allowing real-time tracking of ultra-fast dynamics in quantum materials. This approach will enable the study of transient states, spin dynamics, and quantum coherence under external stimuli.

physics.ins-det↗

arXiv2025

Advanced muon-spin spectroscopy with high lateral resolution using Si-pixel detectors

Muon-spin spectroscopy at continuous sources has stagnated at a stopped muons rate of ~40 kHz for the last few decades. The major limiting factor is the requirement of a single muon in the sample during the typical 10 μs data gate window. To overcome this limit and to be able to perform muon-spin relaxation (μSR) measurements on millimeter-sized samples, one can use vertex reconstruction methods to construct μSR spectra. This is now possible thanks to the availability of very thin monolithic Si-pixel chips, which offer minimal particle scattering and high count rate. Here we present results from a Si-pixel based spectrometer that utilizes vertex reconstruction schemes for the incoming muons and emitted positrons. With this spectrometer we were able to obtain a first vertex reconstructed μSR (VR-μSR) spectrum. The unique capabilities and benefits of such a spectrometer are discussed.

physics.ins-det↗

arXiv2024

Predicting NOx emissions in Biochar Production Plants using Machine Learning

The global Biochar Industry has witnessed a surge in biochar production, with a total of 350k mt/year production in 2023. With the pressing climate goals set and the potential of Biochar Carbon Removal (BCR) as a climate-relevant technology, scaling up the number of new plants to over 1000 facilities per year by 2030 becomes imperative. However, such a massive scale-up presents not only technical challenges but also control and regulation issues, ensuring maximal output of plants while conforming to regulatory requirements. In this paper, we present a novel method of optimizing the process of a biochar plant based on machine learning methods. We show how a standard Random Forest Regressor can be used to model the states of the pyrolysis machine, the physics of which remains highly complex. This model then serves as a surrogate of the machine -- reproducing several key outcomes of the machine -- in a numerical optimization. This, in turn, could enable us to reduce NOx emissions -- a key regulatory goal in that industry -- while achieving maximal output still. In a preliminary test our approach shows remarkable results, proves to be applicable on two different machines from different manufacturers, and can be implemented on standard Internet of Things (IoT) devices more generally.

eess.SP↗

arXiv2024

Can machine learning solve the challenge of adaptive learning and the individualization of learning paths? A field experiment in an online learning platform

The individualization of learning contents based on digital technologies promises large individual and social benefits. However, it remains an open question how this individualization can be implemented. To tackle this question we conduct a randomized controlled trial on a large digital self-learning platform. We develop an algorithm based on two convolutional neural networks that assigns tasks to $4,365$ learners according to their learning paths. Learners are randomized into three groups: two treatment groups -- a group-based adaptive treatment group and an individual adaptive treatment group -- and one control group. We analyze the difference between the three groups with respect to effort learners provide and their performance on the platform. Our null results shed light on the multiple challenges associated with the individualization of learning paths.

cs.LG↗

arXiv2024

10 Years of Fair Representations: Challenges and Opportunities

Fair Representation Learning (FRL) is a broad set of techniques, mostly based on neural networks, that seeks to learn new representations of data in which sensitive or undesired information has been removed. Methodologically, FRL was pioneered by Richard Zemel et al. about ten years ago. The basic concepts, objectives and evaluation strategies for FRL methodologies remain unchanged to this day. In this paper, we look back at the first ten years of FRL by i) revisiting its theoretical standing in light of recent work in deep learning theory that shows the hardness of removing information in neural network representations and ii) presenting the results of a massive experimentation (225.000 model fits and 110.000 AutoML fits) we conducted with the objective of improving on the common evaluation scenario for FRL. More specifically, we use automated machine learning (AutoML) to adversarially "mine" sensitive information from supposedly fair representations. Our theoretical and experimental analysis suggests that deterministic, unquantized FRL methodologies have serious issues in removing sensitive information, which is especially troubling as they might seem "fair" at first glance.

cs.LG↗

arXiv2024

Google Topics as a way out of the cookie dilemma?

The paper discusses the legal requirements and implications of the processing of information and personal data for advertising purposes, particularly in the light of the "Planet49" decision of the European Court of Justice (ECJ) and the "Cookie Consent II" decision by the German Federal Court (Bundesgerichtshof, BGH). It emphasises that obtaining explicit consent of individuals is necessary for setting cookies. The introduction of the German Telecommunication Telemedia Data Protection Act (Telekommunikation-Telemedien-Datenschutzgesetz, TTDSG) has replaced the relevant section of the German Telemedia Act (Telemediengesetz, TMG) and transpose the concept of informed consent for storing and accessing information on terminal equipment, aligning with Article 5(3) ePrivacy Directive. To meet these requirements, companies exploring alternatives to obtaining consent are developing technical mechanisms that rely on a legal basis. Google tested initially "Federated Learning of Cohorts" (FLoC) as part of their "Privacy Sandbox" strategy. This technology was significantly criticized, Google introduced a new project called "Google Topics", which aims to personalize advertising by categorizing users into interest groups, called topics. Implementation of this technology began in July 2023.

cs.CY↗

arXiv2023

Data Flow in the Mu3e DAQ

The Mu3e experiment at the Paul Scherrer Institute (PSI) searches for the charged lepton flavour violating decay $μ^+ \rightarrow e^+ e^+ e^-$. The experiment aims for an ultimate sensitivity of one in $10^{16}$ $μ$ decays. The first phase of the experiment, currently under construction, will reach a branching ratio sensitivity of $2\cdot10^{-15}$ by observing $10^{8}$ $μ$ decays per second over a year of data taking. The highly granular detector based on thin high-voltage monolithic active pixel sensors (HV-MAPS) and scintillating timing detectors will produce about 100 GB/s of data at these particle rates. The Field Programmable Gate Array based Mu3e Data Acquisition System (DAQ) will read out the different detector parts. The trigger-less readout system is used to sort, time align and analyze the data while running. A farm of PCs equipped with powerful graphics processing units (GPUs) will perform the event reconstruction and data reduction. The paper presents the ongoing integration of the sub detectors into the DAQ, especially focusing on the time aligning and the data flow inside the FPGAs of the filter farm. It will also show the DAQ system used in the Mu3e Integration Runs performed in spring 2021 and 2022.

physics.ins-det↗

arXiv2022

Invariant Representations with Stochastically Quantized Neural Networks

Representation learning algorithms offer the opportunity to learn invariant representations of the input data with regard to nuisance factors. Many authors have leveraged such strategies to learn fair representations, i.e., vectors where information about sensitive attributes is removed. These methods are attractive as they may be interpreted as minimizing the mutual information between a neural layer's activations and a sensitive attribute. However, the theoretical grounding of such methods relies either on the computation of infinitely accurate adversaries or on minimizing a variational upper bound of a mutual information estimate. In this paper, we propose a methodology for direct computation of the mutual information between a neural layer and a sensitive attribute. We employ stochastically-activated binary neural networks, which lets us treat neurons as random variables. We are then able to compute (not bound) the mutual information between a layer and a sensitive attribute and use this information as a regularization factor during gradient descent. We show that this method compares favorably with the state of the art in fair representation learning and that the learned representations display a higher level of invariance compared to full-precision neural networks.

cs.LG↗

arXiv2022

Mu3e Integration Run 2021

The Mu3e experiment at the Paul Scherrer Institute searches for the charged lepton flavor violating decay $μ^+ \rightarrow e^+ e^+ e^-$. The experiment aims for an ultimate sensitivity of one in $10^{16}$ $μ$ decays. The first phase of the experiment, currently under construction, will reach a branching ratio sensitivity of $2.5\times10^{-15}$ by observing $10^{8}$ $μ$ decays per second over a year of data taking. The highly granular detector based on thin high-voltage monolithic active pixel sensors (HV-MAPS) and scintillating timing detectors will produce about 100 GB/s of data at these particle rates. The Field Programmable Gate Array (FPGA) based Mu3e Data Acquisition System will read out this data from the detector and reduce the event rate to 100 MB/s by selecting interesting events using a filter farm of graphics processing units. The paper presents the status of the data acquisition system (DAQ) and first results from the 2021 integration run, which for the first time operated a slice of the Mu3e detector at the $π$E5 muon beam line at the Paul Scherrer Institute (PSI).

physics.ins-det↗

arXiv2022

Fair Interpretable Representation Learning with Correction Vectors

Neural network architectures have been extensively employed in the fair representation learning setting, where the objective is to learn a new representation for a given vector which is independent of sensitive information. Various representation debiasing techniques have been proposed in the literature. However, as neural networks are inherently opaque, these methods are hard to comprehend, which limits their usefulness. We propose a new framework for fair representation learning that is centered around the learning of "correction vectors", which have the same dimensionality as the given data vectors. Correction vectors may be computed either explicitly via architectural constraints or implicitly by training an invertible model based on Normalizing Flows. We show experimentally that several fair representation learning models constrained in such a way do not exhibit losses in ranking or classification performance. Furthermore, we demonstrate that state-of-the-art results can be achieved by the invertible model. Finally, we discuss the law standing of our methodology in light of recent legislation in the European Union.

cs.LG↗

arXiv2022

Fair Group-Shared Representations with Normalizing Flows

The issue of fairness in machine learning stems from the fact that historical data often displays biases against specific groups of people which have been underprivileged in the recent past, or still are. In this context, one of the possible approaches is to employ fair representation learning algorithms which are able to remove biases from data, making groups statistically indistinguishable. In this paper, we instead develop a fair representation learning algorithm which is able to map individuals belonging to different groups in a single group. This is made possible by training a pair of Normalizing Flow models and constraining them to not remove information about the ground truth by training a ranking or classification model on top of them. The overall, ``chained'' model is invertible and has a tractable Jacobian, which allows to relate together the probability densities for different groups and ``translate'' individuals from one group to another. We show experimentally that our methodology is competitive with other fair representation learning algorithms. Furthermore, our algorithm achieves stronger invariance w.r.t. the sensitive attribute.

cs.LG↗

arXiv2022

Fair Interpretable Learning via Correction Vectors

Neural network architectures have been extensively employed in the fair representation learning setting, where the objective is to learn a new representation for a given vector which is independent of sensitive information. Various "representation debiasing" techniques have been proposed in the literature. However, as neural networks are inherently opaque, these methods are hard to comprehend, which limits their usefulness. We propose a new framework for fair representation learning which is centered around the learning of "correction vectors", which have the same dimensionality as the given data vectors. The corrections are then simply summed up to the original features, and can therefore be analyzed as an explicit penalty or bonus to each feature. We show experimentally that a fair representation learning problem constrained in such a way does not impact performance.

cs.LG↗

arXiv2021

The Mu3e Data Acquisition

The Mu3e experiment aims to find or exclude the lepton flavour violating decay $μ^+\to e^+e^-e^+$ with a sensitivity of one in 10$^{16}$ muon decays. The first phase of the experiment is currently under construction at the Paul Scherrer Institute (PSI, Switzerland), where beams with up to 10$^8$ muons per second are available. The detector will consist of an ultra-thin pixel tracker made from High-Voltage Monolithic Active Pixel Sensors (HV-MAPS), complemented by scintillating tiles and fibres for precise timing measurements. The experiment produces about 100 Gbit/s of zero-suppressed data which are transported to a filter farm using a network of FPGAs and fast optical links. On the filter farm, tracks and three-particle vertices are reconstructed using highly parallel algorithms running on graphics processing units, leading to a reduction of the data to 100 Mbyte/s for mass storage and offline analysis. The paper introduces the system design and hardware implementation of the Mu3e data acquisition and filter farm.

physics.ins-det↗

arXiv2020

Ranking Creative Language Characteristics in Small Data Scenarios

The ability to rank creative natural language provides an important general tool for downstream language understanding and generation. However, current deep ranking models require substantial amounts of labeled data that are difficult and expensive to obtain for different domains, languages and creative characteristics. A recent neural approach, the DirectRanker, promises to reduce the amount of training data needed but its application to text isn't fully explored. We therefore adapt the DirectRanker to provide a new deep model for ranking creative language with small data. We compare DirectRanker with a Bayesian approach, Gaussian process preference learning (GPPL), which has previously been shown to work well with sparse data. Our experiments with sparse training data show that while the performance of standard neural ranking approaches collapses with small training datasets, DirectRanker remains effective. We find that combining DirectRanker with GPPL increases performance across different settings by leveraging the complementary benefits of both models. Our combined approach outperforms the previous state-of-the-art on humor and metaphor novelty tasks, increasing Spearman's $ρ$ by 14% and 16% on average.

cs.CL↗

arXiv2019

Pairwise Learning to Rank by Neural Networks Revisited: Reconstruction, Theoretical Analysis and Practical Performance

We present a pairwise learning to rank approach based on a neural net, called DirectRanker, that generalizes the RankNet architecture. We show mathematically that our model is reflexive, antisymmetric, and transitive allowing for simplified training and improved performance. Experimental results on the LETOR MSLR-WEB10K, MQ2007 and MQ2008 datasets show that our model outperforms numerous state-of-the-art methods, while being inherently simpler in structure and using a pairwise approach only.

cs.IR↗