arXiv subjects

Martin Becker

Publications and source records attributed to Martin Becker.

At least 19 recordsLinked to original sources

arXiv2026

CytoBERT: A Foundation Model for Cytometry Data

Cytometry measures the complex characteristics of single cells (e.g., counts and protein expression of immune cells) and is widely used across immunological research and clinical settings. However, cytometry data is highly heterogeneous and unstandardized due to experimental protocols and the choice of measured features. While machine learning methods hold the potential to gain deeper insights into cell biology, these challenges make them difficult to apply and transfer across studies. Recent advances in foundation models can alleviate these issues, but corresponding approaches are still scarce in this field. To address this, we provide CytoBERT, a publicly available, open-source, open-weight foundation model for single-cell cytometry data with variable marker panels. CytoBERT is pretrained in a self-supervised manner on a large-scale cytometry corpus (15 human datasets with heterogeneous marker panels and more than 50 million cells) curated through marker standardization, enabling it to learn transferable inter-marker relationships within cells. Fine-tuning CytoBERT for sample-level classification demonstrates that transfer learning across heterogeneous cytometry datasets is feasible, providing a starting point for scalable, generalizable cytometry analysis. Code is available at GitHub.

cs.LG↗

arXiv2026

Zero-Shot Test-Time Canonicalization using Out-of-Distribution Scoring

Pretrained vision models often misclassify inputs that are rotated, scaled, or sheared, even though these affine transformations leave the object class unchanged. Robustness is usually restored either by building equivariance into the architecture or by retraining with augmentation, both of which require changing or retraining the model. Test-time canonicalization instead leaves the classifier untouched. It undoes the transformation of each input, mapping it to a canonical form near the training distribution before classification. Existing canonicalizers, however, rely on a narrow set of logit-based energy scores and bespoke search procedures, leaving the design space of scoring functions and optimizers unexplored. We reframe canonicalization as out-of-distribution (OOD) detection, which lets any OOD score serve as the energy minimized over transformations. Across benchmarks ranging from handwritten characters and sketches to natural images and 3D point clouds, we systematically evaluate around twenty OOD scores and nine search algorithms, finding that distance-based scores paired with random search and local refinement perform best overall. Because canonicalizing an already-aligned input can hurt accuracy, we add a gated mechanism that transforms an input only when its OOD score indicates this is needed, preserving most in-distribution accuracy while retaining the robustness gains on transformed inputs. Code is available at github.com/johschm/its.

cs.CV↗

arXiv2026

On the non-uniformity of the 2026 FIFA World Cup draw

The group stage of a sports tournament is often made more appealing by introducing additional constraints in the group draw that promote an attractive and balanced group composition. For example, the number of intra-regional group matches is minimised in several World Cups. However, under such constraints, the traditional draw procedure may become non-uniform, meaning that the feasible allocations of the teams into groups are not equally likely to occur. Our paper quantifies this non-uniformity of the 2026 FIFA World Cup draw for the official draw procedure, as well as for 47 reasonable alternatives implied by all permutations of the four pots and two group labelling policies. We show why simulating with a recursive backtracking algorithm is intractable, and propose a workable implementation using integer programming. The official draw mechanism is found to be optimal based on four measures of non-uniformity. Nonetheless, non-uniformity can be more than halved if the organiser aims to treat the best teams drawn from the first pot equally.

stat.AP↗

arXiv2025

Informed, but Not Always Improved: Challenging the Benefit of Background Knowledge in GNNs

In complex and low-data domains such as biomedical research, incorporating background knowledge (BK) graphs, such as protein-protein interaction (PPI) networks, into graph-based machine learning pipelines is a promising research direction. However, while BK is often assumed to improve model performance, its actual contribution and the impact of imperfect knowledge remain poorly understood. In this work, we investigate the role of BK in an important real-world task: cancer subtype classification. Surprisingly, we find that (i) state-of-the-art GNNs using BK perform no better than uninformed models like linear regression, and (ii) their performance remains largely unchanged even when the BK graph is heavily perturbed. To understand these unexpected results, we introduce an evaluation framework, which employs (i) a synthetic setting where the BK is clearly informative and (ii) a set of perturbations that simulate various imperfections in BK graphs. With this, we test the robustness of BK-aware models in both synthetic and real-world biomedical settings. Our findings reveal that careful alignment of GNN architectures and BK characteristics is necessary but holds the potential for significant performance improvements.

cs.LG↗

arXiv2025

SubROC: AUC-Based Discovery of Exceptional Subgroup Performance for Binary Classifiers

Machine learning (ML) is increasingly employed in real-world applications like medicine or economics, thus, potentially affecting large populations. However, ML models often do not perform homogeneously, leading to underperformance or, conversely, unusually high performance in certain subgroups (e.g., sex=female AND marital_status=married). Identifying such subgroups can support practical decisions on which subpopulation a model is safe to deploy or where more training data is required. However, an efficient and coherent framework for effective search is missing. Consequently, we introduce SubROC, an open-source, easy-to-use framework based on Exceptional Model Mining for reliably and efficiently finding strengths and weaknesses of classification models in the form of interpretable population subgroups. SubROC incorporates common evaluation measures (ROC and PR AUC), efficient search space pruning for fast exhaustive subgroup search, control for class imbalance, adjustment for redundant patterns, and significance testing. We illustrate the practical benefits of SubROC in case studies as well as in comparative analyses across multiple datasets.

cs.LG↗

arXiv2024

Evolving Markov Chains: Unsupervised Mode Discovery and Recognition from Data Streams

Markov chains are simple yet powerful mathematical structures to model temporally dependent processes. They generally assume stationary data, i.e., fixed transition probabilities between observations/states. However, live, real-world processes, like in the context of activity tracking, biological time series, or industrial monitoring, often switch behavior over time. Such behavior switches can be modeled as transitions between higher-level \emph{modes} (e.g., running, walking, etc.). Yet all modes are usually not previously known, often exhibit vastly differing transition probabilities, and can switch unpredictably. Thus, to track behavior changes of live, real-world processes, this study proposes an online and efficient method to construct Evolving Markov chains (EMCs). EMCs adaptively track transition probabilities, automatically discover modes, and detect mode switches in an online manner. In contrast to previous work, EMCs are of arbitrary order, the proposed update scheme does not rely on tracking windows, only updates the relevant region of the probability tensor, and enjoys geometric convergence of the expected estimates. Our evaluation of synthetic data and real-world applications on human activity recognition, electric motor condition monitoring, and eye-state recognition from electroencephalography (EEG) measurements illustrates the versatility of the approach and points to the potential of EMCs to efficiently track, model, and understand live, real-world processes.

cs.LG↗

arXiv2021

Redescription Model Mining

This paper introduces Redescription Model Mining, a novel approach to identify interpretable patterns across two datasets that share only a subset of attributes and have no common instances. In particular, Redescription Model Mining aims to find pairs of describable data subsets -- one for each dataset -- that induce similar exceptional models with respect to a prespecified model class. To achieve this, we combine two previously separate research areas: Exceptional Model Mining and Redescription Mining. For this new problem setting, we develop interestingness measures to select promising patterns, propose efficient algorithms, and demonstrate their potential on synthetic and real-world data. Uncovered patterns can hint at common underlying phenomena that manifest themselves across datasets, enabling the discovery of possible associations between (combinations of) attributes that do not appear in the same dataset.

cs.DB↗

arXiv2021

Thermal tuning of a fiber-integrated Fabry-P\'erot cavity

Here, we present the thermal tuning capability of an alignment-free fiber-integrated Fabry-P\'erot cavity.The two mirrors are made of fiber Bragg gratings that can be individually temperature stabilized and tuned. We show the temperature tuning of the resonance wavelength of the cavity without any degradation of the finesse and the tuning of the individual stop bands of the fiber Bragg gratings. This not only permits for the cavity's finesse to be optimized post-fabrication but also makes this cavity applicable as a narrowband filter with a FWHM spectral width of 0.07 (0.02) pm and a suppression of more than -15 dB that can be wavelength tuned. Further, in the field of quantum optics, where strong light-matter interactions are desirable, quantum emitters can be coupled to such a cavity and the cavity effect can be reversibly omitted and re-established. This is particularly useful when working with solid-state quantum emitters where such a reference measurement is often not possible once an emitter has been permanently deposited inside a cavity.

physics.ins-det↗

arXiv2020

MapLUR: Exploring a new Paradigm for Estimating Air Pollution using Deep Learning on Map Images

Land-use regression (LUR) models are important for the assessment of air pollution concentrations in areas without measurement stations. While many such models exist, they often use manually constructed features based on restricted, locally available data. Thus, they are typically hard to reproduce and challenging to adapt to areas beyond those they have been developed for. In this paper, we advocate a paradigm shift for LUR models: We propose the Data-driven, Open, Global (DOG) paradigm that entails models based on purely data-driven approaches using only openly and globally available data. Progress within this paradigm will alleviate the need for experts to adapt models to the local characteristics of the available data sources and thus facilitate the generalizability of air pollution models to new areas on a global scale. In order to illustrate the feasibility of the DOG paradigm for LUR, we introduce a deep learning model called MapLUR. It is based on a convolutional neural network architecture and is trained exclusively on globally and openly available map data without requiring manual feature engineering. We compare our model to state-of-the-art baselines like linear regression, random forests and multi-layer perceptrons using a large data set of modeled $\text{NO}_2$ concentrations in Central London. Our results show that MapLUR significantly outperforms these approaches even though they are provided with manually tailored features. Furthermore, we illustrate that the automatic feature extraction inherent to models based on the DOG paradigm can learn features that are readily interpretable and closely resemble those commonly used in traditional LUR approaches.

cs.LG↗

arXiv2020

Nanofiber-based high-Q microresonator for cryogenic applications

We demonstrate a cryo-compatible, fully fiber-integrated, alignment-free optical microresonator. The compatibility with low temperatures expands its possible applications to the wide field of solid-state quantum optics, where a cryogenic environment is often a requirement. At a temperature of 4.6 K we obtain a quality factor of $\mathbf{(9.9 \pm 0.7) \times 10^6}$. In conjunction with the small mode volume provided by the nanofiber, this cavity can be either used in the coherent dynamics or the fast cavity regime, where it can provide a Purcell factor of up to 15. Our resonator is therefore suitable for significantly enhancing the coupling between light and a large variety of different quantum emitters and due to its proven performance over a wide temperature range, also lends itself for the implementation of quantum hybrid systems.

quant-ph↗

arXiv2019

A Valgrind Tool to Compute the Working Set of a Software Process

This paper introduces a new open-source tool for the dynamic analyzer Valgrind. The tool measures the amount of memory that is actively being used by a process at any given point in time. While there exist numerous tools to measure the memory requirements of a process, the vast majority only focuses on metrics like resident or proportional set sizes, which include memory that was once claimed, but is momentarily disused. Consequently, such tools do not permit drawing conclusions about how much cache or RAM a process actually requires at each point in time, and thus cannot be used for performance debugging. The few tools which do measure only actively used memory, however, have limitations in temporal resolution and introspection. In contrast, our tool offers an easy way to compute the memory that has recently been accessed at any point in time, reflecting how cache and RAM requirements change over time. In particular, this tool computes the set of memory references made within a fixed time interval before any point in time, known as the working set, and captures call stacks for interesting peaks in the working set size. We first introduce the tool, then we run some examples comparing the output from our tool with similar memory tools, and we close with a discussion of limitations

cs.PF↗

arXiv2018

Measuring Software Performance on Linux

Measuring and analyzing the performance of software has reached a high complexity, caused by more advanced processor designs and the intricate interaction between user programs, the operating system, and the processor's microarchitecture. In this report, we summarize our experience about how performance characteristics of software should be measured when running on a Linux operating system and a modern processor. In particular, (1) We provide a general overview about hardware and operating system features that may have a significant impact on timing and how they interact, (2) we identify sources of errors that need to be controlled in order to obtain unbiased measurement results, and (3) we propose a measurement setup for Linux to minimize errors. Although not the focus of this report, we describe the measurement process using hardware performance counters, which can faithfully reflect the real bottlenecks on a given processor. Our experiments confirm that our measurement setup has a large impact on the results. More surprisingly, however, they also suggest that the setup can be negligible for certain analysis methods. Furthermore, we found that our setup maintains significantly better performance under background load conditions, which means it can be used to improve software in high-performance applications.

cs.PF↗

arXiv2018

Scalable and Precise Estimation and Debugging of the Worst-Case Execution Time for Analysis-Friendly Processors

Estimating the Worst-Case Execution Time (WCET) of an application is an essential task in the context of developing real-time or safety-critical software, but it is also a complex and error-prone process. Conventional approaches require at least some manual inputs from the user, such as loop bounds and infeasible path information, which are hard to obtain and can lead to unsafe results if they are incorrect. This is aggravated by the lack of a comprehensive explanation of the WCET estimate, i.e., a specific trace showing how WCET was reached. It is therefore hard to spot incorrect inputs and hard to improve the worst-case timing of the application. Meanwhile, modern processors have reached a complexity that refutes analysis and puts more and more burden on the practitioner. In this article we show how all of these issues can be significantly mitigated or even solved, if we use processors that are amenable to WCET analysis. We define and identify such processors, and then we propose an automated tool set which estimates a precise WCET without unsafe manual inputs, and also reconstructs a maximum-detail view of the WCET path that can be examined in a debugger environment. Our approach is based on Model Checking, which however is known to scale badly with growing application size. We address this issue by shifting the analysis to source code level, where source code transformations can be applied that retain the timing behavior, but reduce the complexity. Our experiments show that fast and precise estimates can be achieved with Model Checking, that its scalability can even exceed current approaches, and that new opportunities arise in the context of "timing debugging".

cs.SE↗

arXiv2017

Adaptive kNN using Expected Accuracy for Classification of Geo-Spatial Data

The k-Nearest Neighbor (kNN) classification approach is conceptually simple - yet widely applied since it often performs well in practical applications. However, using a global constant k does not always provide an optimal solution, e.g., for datasets with an irregular density distribution of data points. This paper proposes an adaptive kNN classifier where k is chosen dynamically for each instance (point) to be classified, such that the expected accuracy of classification is maximized. We define the expected accuracy as the accuracy of a set of structurally similar observations. An arbitrary similarity function can be used to find these observations. We introduce and evaluate different similarity functions. For the evaluation, we use five different classification tasks based on geo-spatial data. Each classification task consists of (tens of) thousands of items. We demonstrate, that the presented expected accuracy measures can be a good estimator for kNN performance, and the proposed adaptive kNN classifier outperforms common kNN and previously introduced adaptive kNN algorithms. Also, we show that the range of considered k can be significantly reduced to speed up the algorithm without negative influence on classification accuracy.

cs.CV↗

arXiv2017

Development and Verification of a Flight Stack for a High-Altitude Glider in Ada/SPARK 2014

SPARK 2014 is a modern programming language and a new state-of-the-art tool set for development and verification of high-integrity software. In this paper, we explore the capabilities and limitations of its latest version in the context of building a flight stack for a high-altitude unmanned glider. Towards that, we deliberately applied static analysis early and continuously during implementation, to give verification the possibility to steer the software design. In this process we have identified several limitations and pitfalls of software design and verification in SPARK, for which we give workarounds and protective actions to avoid them. Finally, we give design recommendations that have proven effective for verification, and summarize our experiences with this new language.

cs.SE↗

arXiv2017

Learning Semantic Relatedness From Human Feedback Using Metric Learning

Assessing the degree of semantic relatedness between words is an important task with a variety of semantic applications, such as ontology learning for the Semantic Web, semantic search or query expansion. To accomplish this in an automated fashion, many relatedness measures have been proposed. However, most of these metrics only encode information contained in the underlying corpus and thus do not directly model human intuition. To solve this, we propose to utilize a metric learning approach to improve existing semantic relatedness measures by learning from additional information, such as explicit human feedback. For this, we argue to use word embeddings instead of traditional high-dimensional vector representations in order to leverage their semantic density and to reduce computational cost. We rigorously test our approach on several domains including tagging data as well as publicly available embeddings based on Wikipedia texts and navigation. Human feedback about semantic relatedness for learning and evaluation is extracted from publicly available datasets such as MEN or WS-353. We find that our method can significantly improve semantic relatedness measures by learning from additional information, such as explicit human feedback. For tagging data, we are the first to generate and study embeddings. Our results are of special interest for ontology and recommendation engineers, but also for any other researchers and practitioners of Semantic Web techniques.

cs.CL↗

arXiv2016

MixedTrails: Bayesian hypothesis comparison on heterogeneous sequential data

Sequential traces of user data are frequently observed online and offline, e.g., as sequences of visited websites or as sequences of locations captured by GPS. However, understanding factors explaining the production of sequence data is a challenging task, especially since the data generation is often not homogeneous. For example, navigation behavior might change in different phases of browsing a website, or movement behavior may vary between groups of users. In this work, we tackle this task and propose MixedTrails, a Bayesian approach for comparing the plausibility of hypotheses regarding the generative processes of heterogeneous sequence data. Each hypothesis is derived from existing literature, theory or intuition and represents a belief about transition probabilities between a set of states that can vary between groups of observed transitions. For example, when trying to understand human movement in a city and given some observed data, a hypothesis assuming tourists to be more likely to move towards points of interests than locals, can be shown to be more plausible than a hypothesis assuming the opposite. Our approach incorporates such hypotheses as Bayesian priors in a generative mixed transition Markov chain model, and compares their plausibility utilizing Bayes factors. We discuss analytical and approximate inference methods for calculating the marginal likelihoods for Bayes factors, give guidance on interpreting the results, and illustrate our approach with several experiments on synthetic and empirical data from Wikipedia and Flickr. Thus, this work enables a novel kind of analysis for studying sequential data in many application areas.

cs.SI↗

arXiv2016

Analyzing Features for the Detection of Happy Endings in German Novels

With regard to a computational representation of literary plot, this paper looks at the use of sentiment analysis for happy ending detection in German novels. Its focus lies on the investigation of previously proposed sentiment features in order to gain insight about the relevance of specific features on the one hand and the implications of their performance on the other hand. Therefore, we study various partitionings of novels, considering the highly variable concept of "ending". We also show that our approach, even though still rather simple, can potentially lead to substantial findings relevant to literary studies.

cs.IR↗