arXiv subjects

Christoph Schommer

Publications and source records attributed to Christoph Schommer.

At least 19 recordsLinked to original sources

arXiv2026

Feature Generation Using LLMs: An Evolutionary Algorithm Approach

A crucial step in machine learning pipelines is to present each entity with features or attributes that are representative of the characteristics of the processed entities. Feature engineering is an important step in finding a relation among attributes that otherwise may not be processed by the ML algorithms. Meanwhile, Large Language Models have shown promising abilities in coding, mathematical reasoning, and processing world knowledge. In this work, we utilize an LLM for the problem of feature generation from tabular data based on the previously given features. We have created a pipeline that takes a set of attributes and a prompt to generate new features. Then, our selection algorithm selects the best-performing sets of attributes. We apply our method to eight datasets from different domains and data types. Our results show that, in most cases, the language model can produce new features based on mathematical and logical operators that are useful for the given tasks and can improve classification results.

cs.LG↗

arXiv2026

AEyeDE: An Attention-Based Attribution Framework for AI-Generated Text Detection

Detecting AI-generated text is becoming increasingly challenging as modern language models approach human-level fluency and can evade detectors that rely on surface statistics or likelihood-based signals. We propose \textsc{AEyeDE}, an attribution-driven approach to human-AI authorship detection that leverages model attention as a discriminative signal. Specifically, we extract attention-based attribution matrices for both human- and AI-generated text using a \emph{proxy} Transformer model with white-box access and train a lightweight Convolutional Neural Network to learn representations from these attribution maps. Across encoder-decoder translation settings, our method consistently outperforms a text-only baseline. In decoder-only settings, it performs strongly in generator-specific detection, remains competitive on standard benchmarks, and shows robustness under cross-dataset transfer and alternative-spelling perturbations. We further show that attention maps exhibit recurring local structures whose relative frequencies differ consistently between human- and AI-generated text across datasets and proxy models. These findings suggest that attention-based attribution maps provide a complementary and interpretable signal for AI-generated text detection. We will make the code publicly available to support future research.

cs.CL↗

arXiv2026

Evaluating Explainable AI Attribution Methods in Neural Machine Translation via Attention-Guided Knowledge Distillation

The study of the attribution of input features to the output of neural network models is an active area of research. While numerous Explainable AI (XAI) techniques have been proposed to interpret these models, the systematic and automated evaluation of these methods in sequence-to-sequence (seq2seq) models is less explored. This paper introduces a new approach for evaluating explainability methods in transformer-based seq2seq models. We use teacher-derived attribution maps as a structured side signal to guide a student model, and quantify the utility of different attribution methods through the student's ability to simulate targets. Using the Inseq library, we extract attribution scores over source-target sequence pairs and inject these scores into the attention mechanism of a student transformer model under four composition operators (addition, multiplication, averaging, and replacement). Across three language pairs (de-en, fr-en, ar-en) and attributions from Marian-MT and mBART models, Attention, Value Zeroing, and Layer Gradient $\times$ Activation consistently yield the largest gains in BLEU (and corresponding improvements in chrF) relative to baselines. In contrast, other gradient-based methods (Saliency, Integrated Gradients, DeepLIFT, Input $\times$ Gradient, GradientShap) lead to smaller and less consistent improvements. These results suggest that different attribution methods capture distinct signals and that attention-derived attributions better capture alignment between source and target representations in seq2seq models. Finally, we introduce an Attributor transformer that, given a source-target pair, learns to reconstruct the teacher's attribution map. Our findings demonstrate that the more accurately the Attributor can reproduce attribution maps, the more useful an injection of those maps is for the downstream task. The source code can be found on GitHub.

cs.CL↗

arXiv2025

Cluster Purge Loss: Structuring Transformer Embeddings for Equivalent Mutants Detection

Recent pre-trained transformer models achieve superior performance in various code processing objectives. However, although effective at optimizing decision boundaries, common approaches for fine-tuning them for downstream classification tasks - distance-based methods or training an additional classification head - often fail to thoroughly structure the embedding space to reflect nuanced intra-class semantic relationships. Equivalent code mutant detection is one of these tasks, where the quality of the embedding space is crucial to the performance of the models. We introduce a novel framework that integrates cross-entropy loss with a deep metric learning objective, termed Cluster Purge Loss. This objective, unlike conventional approaches, concentrates on adjusting fine-grained differences within each class, encouraging the separation of instances based on semantical equivalency to the class center using dynamically adjusted borders. Employing UniXCoder as the base model, our approach demonstrates state-of-the-art performance in the domain of equivalent mutant detection and produces a more interpretable embedding space.

cs.LG↗

arXiv2023

Compatibility of Fairness Metrics with EU Non-Discrimination Laws: Demographic Parity & Conditional Demographic Disparity

Empirical evidence suggests that algorithmic decisions driven by Machine Learning (ML) techniques threaten to discriminate against legally protected groups or create new sources of unfairness. This work supports the contextual approach to fairness in EU non-discrimination legal framework and aims at assessing up to what point we can assure legal fairness through fairness metrics and under fairness constraints. For that, we analyze the legal notion of non-discrimination and differential treatment with the fairness definition Demographic Parity (DP) through Conditional Demographic Disparity (CDD). We train and compare different classifiers with fairness constraints to assess whether it is possible to reduce bias in the prediction while enabling the contextual approach to judicial interpretation practiced under EU non-discrimination laws. Our experimental results on three scenarios show that the in-processing bias mitigation algorithm leads to different performances in each of them. Our experiments and analysis suggest that AI-assisted decision-making can be fair from a legal perspective depending on the case at hand and the legal justification. These preliminary results encourage future work which will involve further case studies, metrics, and fairness notions.

cs.CY↗

arXiv2009

A Cognitive Mind-map Framework to Foster Trust

The explorative mind-map is a dynamic framework, that emerges automatically from the input, it gets. It is unlike a verificative modeling system where existing (human) thoughts are placed and connected together. In this regard, explorative mind-maps change their size continuously, being adaptive with connectionist cells inside; mind-maps process data input incrementally and offer lots of possibilities to interact with the user through an appropriate communication interface. With respect to a cognitive motivated situation like a conversation between partners, mind-maps become interesting as they are able to process stimulating signals whenever they occur. If these signals are close to an own understanding of the world, then the conversational partner becomes automatically more trustful than if the signals do not or less match the own knowledge scheme. In this (position) paper, we therefore motivate explorative mind-maps as a cognitive engine and propose these as a decision support engine to foster trust.

cs.AI↗

arXiv2009

Symbolic Computing with Incremental Mindmaps to Manage and Mine Data Streams - Some Applications

In our understanding, a mind-map is an adaptive engine that basically works incrementally on the fundament of existing transactional streams. Generally, mind-maps consist of symbolic cells that are connected with each other and that become either stronger or weaker depending on the transactional stream. Based on the underlying biologic principle, these symbolic cells and their connections as well may adaptively survive or die, forming different cell agglomerates of arbitrary size. In this work, we intend to prove mind-maps' eligibility following diverse application scenarios, for example being an underlying management system to represent normal and abnormal traffic behaviour in computer networks, supporting the detection of the user behaviour within search engines, or being a hidden communication layer for natural language interaction.

cs.NE↗

arXiv2008

CoZo+ - A Content Zoning Engine for textual documents

Content zoning can be understood as a segmentation of textual documents into zones. This is inspired by [6] who initially proposed an approach for the argumentative zoning of textual documents. With the prototypical CoZo+ engine, we focus on content zoning towards an automatic processing of textual streams while considering only the actors as the zones. We gain information that can be used to realize an automatic recognition of content for pre-defined actors. We understand CoZo+ as a necessary pre-step towards an automatic generation of summaries and to make intellectual ownership of documents detectable.

cs.CL↗

arXiv2008

Assembling Actor-based Mind-Maps from Text Stream

For human beings, the processing of text streams of unknown size leads generally to problems because e.g. noise must be selected out, information be tested for its relevance or redundancy, and linguistic phenomenon like ambiguity or the resolution of pronouns be advanced. Putting this into simulation by using an artificial mind-map is a challenge, which offers the gate for a wide field of applications like automatic text summarization or punctual retrieval. In this work we present a framework that is a first step towards an automatic intellect. It aims at assembling a mind-map based on incoming text streams and on a subject-verb-object strategy, having the verb as an interconnection between the adjacent nouns. The mind-map's performance is enriched by a pronoun resolution engine that bases on the work of D. Klein, and C. D. Manning.

cs.CL↗

arXiv2008

16 Propositions to Reconsider the Organization of a Scientific Workshop

Participating a scientific workshop is nowadays often an adventure because the number of participants do seldom exceed the number of talks. A half-day workshop is mostly finished at lunchtime, speakers are sometimes not present and unexcused, and a strict progression of the workshop offers little air for discussion. And when talks are re-scheduled on short notice in case that a speech is dropped out, attaining guests definitely wonder why the presenter is talking about something that does not match the previously announced talk. In this respect, we believe that the organization of a workshop in the classical sense must be reconsidered. It is not enough of compelling the presenters to pay the registration fee only and to let the participants being impassive or taken away mentally. With this work, we address several propositions to become implemented in the future workshop organization. With that, we hope to contribute to the identification of scientific workshops as a place of interaction.

cs.CY↗

arXiv2008

An Unified Definition of Data Mining

Since many years, theoretical concepts of Data Mining have been developed and improved. Data Mining has become applied to many academic and industrial situations, and recently, soundings of public opinion about privacy have been carried out. However, a consistent and standardized definition is still missing, and the initial explanation given by Frawley et al. has pragmatically often changed over the years. Furthermore, alternative terms like Knowledge Discovery have been conjured and forged, and a necessity of a Data Warehouse has been endeavoured to persuade the users. In this work, we pick up current definitions and introduce an unified definition that covers existing attempted explanations. For this, we appeal to the natural original of chemical states of aggregation.

cs.SC↗

arXiv2008

A Simple Framework to Typify Social Bibliographic Communities

Social Communities in bibliographic databases exist since many years, researchers share common research interests, and work and publish together. A social community may vary in type and size, being fully connected between participating members or even more expressed by a consortium of small and individual members who play individual roles in it. In this work, we focus on social communities inside the bibliographic database DBLP and characterize communities through a simple typifying description model. Generally, we understand a publication as a transaction between the associated authors. The idea therefore is to concern with directed associative relationships among them, to decompose each pattern to its fundamental structure, and to describe the communities by expressive attributes. Finally, we argue that the decomposition supports the management of discovered structures towards the use of adaptive-incremental mind-maps.

cs.DL↗

arXiv2008

A Network Protection Framework through Artificial Immunity

Current network protection systems use a collection of intelligent components - e.g. classifiers or rule-based firewall systems to detect intrusions and anomalies and to secure a network against viruses, worms, or trojans. However, these network systems rely on individuality and support an architecture with less collaborative work of the protection components. They give less administration support for maintenance, but offer a large number of individual single points of failures - an ideal situation for network attacks to succeed. In this work, we discuss the required features, the performance, and the problems of a distributed protection system called {\it SANA}. It consists of a cooperative architecture, it is motivated by the human immune system, where the components correspond to artificial immune cells that are connected for their collaborative work. SANA promises a better protection against intruders than common known protection systems through an adaptive self-management while keeping the resources efficiently by an intelligent reduction of redundancies. We introduce a library of several novel and common used protection components and evaluate the performance of SANA by a proof-of-concept implementation.

cs.MA↗

arXiv2008

A Simple Dynamic Mind-map Framework To Discover Associative Relationships in Transactional Data Streams

In this paper, we informally introduce dynamic mind-maps that represent a new approach on the basis of a dynamic construction of connectionist structures during the processing of a data stream. This allows the representation and processing of recursively defined structures and avoids the problem of a more traditional, fixed-size architecture with the processing of input structures of unknown size. For a data stream analysis with association discovery, the incremental analysis of data leads to results on demand. Here, we describe a framework that uses symbolic cells to calculate associations based on transactional data streams as it exists in e.g. bibliographic databases. We follow a natural paradigm of applying simple operations on cells yielding on a mind-map structure that adapts over time.

cs.NE↗

arXiv2008

A new queueing strategy for the Adversarial Queueing Theory

In the today's Internet and TCP/IP-networks, the queueing of packets is commonly implemented using the protocol FIFO (First In First Out). Unfortunately, FIFO performs poorly in the Adversarial Queueing Theory. Other queueing strategies are researched in this model and better results are performed by alternative queueing strategies, e.g. LIS (Longest In System). This article introduces a new queueing protocol called interval-strategy that is concerned with the reduction from dynamic to static routing. We discuss the maximum system time for a packet and estimate with up-to-date results how this can be achieved. We figure out the maximum amount of time where a packet can spend in the network (i.e. worst case system time), and argue that the universal instability of the presented interval-strategy can be reached through these results. When a large group of queueing strategies is used for queueing, we prove that the interval-strategy will be universally unstable. Finally, we calculate the maximum time of the static routing to reach an universal stable and polynomial - in detail linear - bounded interval-strategy. Afterwards we close - in order to check this upper bound - with up-to-date results about the delivery times in static routing.

cs.NI↗

arXiv2008

AGNOSCO - Identification of Infected Nodes with artificial Ant Colonies

If a computer node is infected by a virus, worm or a backdoor, then this is a security risk for the complete network structure where the node is associated. Existing Network Intrusion Detection Systems (NIDS) provide a certain amount of support for the identification of such infected nodes but suffer from the need of plenty of communication and computational power. In this article, we present a novel approach called AGNOSCO to support the identification of infected nodes through the usage of artificial ant colonies. It is shown that AGNOSCO overcomes the communication and computational power problem while identifying infected nodes properly.

cs.AI↗

arXiv2008

SANA - Network Protection through artificial Immunity

Current network protection systems use a collection of intelligent components - e.g. classifiers or rule-based firewall systems to detect intrusions and anomalies and to secure a network against viruses, worms, or trojans. However, these network systems rely on individuality and support an architecture with less collaborative work of the protection components. They give less administration support for maintenance, but offer a large number of individual single points of failures - an ideal situation for network attacks to succeed. In this work, we discuss the required features, the performance, and the problems of a distributed protection system called SANA. It consists of a cooperative architecture, it is motivated by the human immune system, where the components correspond to artificial immune cells that are connected for their collaborative work. SANA promises a better protection against intruders than common known protection systems through an adaptive self-management while keeping the resources efficiently by an intelligent reduction of redundant tasks. We introduce a library of several novel and common used protection components and evaluate the performance of SANA by a proof-of-concept implementation.

cs.CR↗

arXiv2008

Service Oriented Architecture in Network Security - a novel Organisation in Security Systems

Current network security systems are a collection of various security components, which are directly installed in the operating system. These check the whole node for suspicious behaviour. Armouring intrusions e.g. have the ability to hide themselves from being checked. We present in this paper an alternative organisation of security systems. The node is completely virtualized with current virtualization systems so that the operating system with applications and the security system is distinguished. The security system then checks the node from outside and the right security components are provided through a service oriented architecture. Due to the running in a virtual machine, the infected nodes can be halted, duplicated, and moved to other nodes for further analysis and legal aspects. This organisation is in this article analysed and a preliminary implementation showing promising results are discussed.

cs.CR↗