arXiv subjects

Andreas Henschel

Publications and source records attributed to Andreas Henschel.

7 recordsLinked to original sources

arXiv2026

DERA: Detached Edge-Residual Adaptation for Prohibited item Detection

Prohibited-item detection in X-ray imagery remains challenging due to object superposition, weak texture, and material clutter which obscure both semantic appearance and object boundaries. We propose \textbf{DERA}, a \textbf{D}etached \textbf{E}dge-\textbf{R}esidual \textbf{A}daptation framework for prohibited item detection under X-ray imagery. DERA combines hierarchical visual features with a parallel pixel-difference edge pyramid and learns an object-specific boundary prior from training-time contours of the instance masks. The detached prior gates edge-sensitive features, which are injected into the early visual stages through residual heads. This staged design preserves the foundation detector at the start of adaptation, isolates boundary supervision from semantic feature learning, and restricts the final adaptation stage to only \(14.7\)K trainable parameters. Evaluated on PIDray, CLCXray, and STCray, DERA improves the baseline by \textbf{3.1}, \textbf{1.6}, and \textbf{2.4} AP points, respectively.

cs.CV↗

arXiv2026

Falcon: Functional Assembly and Language for Compositional Reasoning in X-ray

Conventional vision-language models are largely object-centric, focusing on detecting and describing individual entities. In safety-critical X-ray baggage screening, however, threat often emerges not from a single object but from the functional compatibility of spatially dispersed components, such as batteries, detonators, and explosive charges. We formalize this setting as \emph{compositional threat reasoning}, where risk is modeled as a relational property of grounded regions rather than an independent detection outcome. We introduce \textbf{Falcon}, a multimodal framework that abstracts segmentation-aware region features into a structured safety state capturing component presence, pairwise functional compatibility, and scene-level risk. This structured representation is injected into the language model as an explicit intermediate interface, encouraging relationally consistent and safety-aware reasoning. To evaluate this problem, we present \textbf{Falcon-X}, a benchmark that unifies dense grounding with structured supervision over component completeness and risk inference in cluttered X-ray imagery. Experiments show that while existing multimodal models adapt to appearance, they struggle with compositional safety reasoning. Falcon improves functional grounding and produces more coherent threat assessments, establishing compositional safety reasoning as a distinct evaluation paradigm for multimodal systems.

cs.CV↗

arXiv2026

Benchmarking end-to-end genotype-to-phenotype prediction workflows across 80 openSNP phenotypes

Genotype-to-phenotype prediction is a central goal of statistical genetics, yet practical comparisons of prediction workflows remain limited in small, heterogeneous, participant-shared genomic datasets. Here, we benchmarked end-to-end case-control prediction across 80 curated binary phenotypes from openSNP using machine learning, deep learning, and polygenic score workflows. We evaluated 29 machine-learning algorithms, 80 deep-learning model variants, and 3 polygenic score tools across 675 clumping and pruning configurations. No workflow family dominated universally. Polygenic score workflows achieved the highest observed discrimination for 53 phenotypes, whereas machine-learning or deep-learning workflows achieved the highest for 27. However, many apparent phenotype-level wins were modest, with 41.2\% of comparisons representing practical ties within five discrimination points. Performance was strongly phenotype-dependent and sensitive to modeling and preprocessing choices. Distinct workflow-specific failure modes were also observed, including unstable behaviour in PRSice and non-informative collapse in lassosum for 13 phenotypes. Higher peak performance was concentrated in smaller phenotypes, reinforcing the need for cautious interpretation in limited-data settings. The cohort was predominantly of European ancestry, restricting generalisability. Together, these results position openSNP as a useful stress-test environment for genomic prediction and support benchmark-guided workflow selection under realistic conditions of data scarcity, phenotype heterogeneity, and ancestry imbalance.

q-bio.GN↗

arXiv2024

Mapping Hong Kong's Financial Ecosystem: A Network Analysis of the SFC's Licensed Professionals and Institutions

We present the first study of the Public Register of Licensed Persons and Registered Institutions maintained by the Hong Kong Securities and Futures Commission (SFC) through the lens of complex network analysis. This dataset, spanning 21 years with daily granularity, provides a unique view of the evolving social network between licensed professionals and their affiliated firms in Hong Kong's financial sector. Leveraging large language models, we classify firms (e.g., asset managers, banks) and infer the likely nationality and gender of employees based on their names. This application enhances the dataset by adding rich demographic and organizational context, enabling more precise network analysis. Our preliminary findings reveal key structural features, offering new insights into the dynamics of Hong Kong's financial landscape. We release the structured dataset to enable further research, establishing a foundation for future studies that may inform recruitment strategies, policy-making, and risk management in the financial industry.

stat.AP↗

arXiv2024

Enriching Datasets with Demographics through Large Language Models: What's in a Name?

Enriching datasets with demographic information, such as gender, race, and age from names, is a critical task in fields like healthcare, public policy, and social sciences. Such demographic insights allow for more precise and effective engagement with target populations. Despite previous efforts employing hidden Markov models and recurrent neural networks to predict demographics from names, significant limitations persist: the lack of large-scale, well-curated, unbiased, publicly available datasets, and the lack of an approach robust across datasets. This scarcity has hindered the development of traditional supervised learning approaches. In this paper, we demonstrate that the zero-shot capabilities of Large Language Models (LLMs) can perform as well as, if not better than, bespoke models trained on specialized data. We apply these LLMs to a variety of datasets, including a real-life, unlabelled dataset of licensed financial professionals in Hong Kong, and critically assess the inherent demographic biases in these models. Our work not only advances the state-of-the-art in demographic enrichment but also opens avenues for future research in mitigating biases in LLMs.

cs.CL↗

arXiv2024

Informational Rescaling of PCA Maps with Application to Genetic Distance

We discuss the inadequacy of covariances/correlations and other measures in L2 as relative distance metrics under some conditions. We propose a computationally simple heuristic to transform a map based on standard principal component analysis (PCA) (when the variables are asymptotically Gaussian) into an entropy-based map where distances are based on mutual information (MI). Rescaling Principal Component based distances using MI allows a representation of relative statistical associations when, as in genetics, it is applied on bit measurements between individuals' genomic mutual information. This entropy rescaled PCA, while preserving order relationships (along a dimension), changes the relative distances to make them linear to information. We show the effect on the entire world population and some subsamples, which leads to significant differences with the results of current research.

cs.IT↗

arXiv2022

Deep learning pipeline for image classification on mobile phones

This article proposes and documents a machine-learning framework and tutorial for classifying images using mobile phones. Compared to computers, the performance of deep learning model performance degrades when deployed on a mobile phone and requires a systematic approach to find a model that performs optimally on both computers and mobile phones. By following the proposed pipeline, which consists of various computational tools, simple procedural recipes, and technical considerations, one can bring the power of deep learning medical image classification to mobile devices, potentially unlocking new domains of applications. The pipeline is demonstrated on four different publicly available datasets: COVID X-rays, COVID CT scans, leaves, and colorectal cancer. We used two application development frameworks: TensorFlow Lite (real-time testing) and Flutter (digital image testing) to test the proposed pipeline. We found that transferring deep learning models to a mobile phone is limited by hardware and classification accuracy drops. To address this issue, we proposed this pipeline to find an optimized model for mobile phones. Finally, we discuss additional applications and computational concerns related to deploying deep-learning models on phones, including real-time analysis and image preprocessing. We believe the associated documentation and code can help physicians and medical experts develop medical image classification applications for distribution.

eess.IV↗