arXiv subjects

Jörg Schlötterer

Publications and source records attributed to Jörg Schlötterer.

At least 19 recordsLinked to original sources

arXiv2026

Targeted Visual Counterfactual Explanations for Contrastive Vision-Language Model

Current explanation methods for contrastive vision-language models such as CLIP mainly identify important regions without showing how to change the input in order to get a target prediction. We introduce Mask-guided Adaptive Counterfactual Explanations (MACE), a targeted visual counterfactual method designed specifically for CLIP zero-shot classification. MACE constructs an editable region from either source attribution or source-target attribution differences and expands the mask only when needed to reach a specified target class. A latent diffusion inpainting model then modifies the selected region, while a frozen CLIP model provides modification guidance and anchors the remaining image content to the original input. We evaluate MACE on ImageNet, Food-101, Oxford Pets, and CUB-200. The source-mask variant achieves the highest target top-1 success rate across all four datasets, while the difference-mask variant produces the smallest pixel level and perceptual changes and the best realism scores. Both variants improve proximity and realism over a Stable Diffusion-only baseline using the same generative backbone. These results show that adaptive mask-guided editing produces effective CLIP counterfactuals. They further reveal a tradeoff between counterfactual validity and source-image preservation.

cs.CV↗

arXiv2026

Can Fine-Tuning Erase Your Edits? On the Fragile Coexistence of Knowledge Editing and Adaptation

Knowledge editing (KE) offers a lightweight alternative to retraining for updating large language models (LLMs). Meanwhile, fine-tuning remains the default operation for adapting LLMs to new domains and tasks. Despite their widespread adoption, these two post-training interventions have been studied in isolation, leaving open a crucial question: if we fine-tune an edited model, do the edits survive? This question is motivated by practical objectives: removing covert or malicious edits, and preserving beneficial edits. If fine-tuning impairs edits (Fig.1), current KE methods become less efficient, as a newly fine-tuned model requires re-editing; if edits persist, fine-tuned models risk propagating hidden malicious edits, raising serious safety concerns. To this end, we systematically quantify edit decay after fine-tuning across 254 experimental configurations. Our results show that in general, edits decay substantially after subsequent fine-tuning. AlphaEdit exhibits the greatest decay on the zsRE benchmark when applied to GPT-J, where 25.27% of previously successful edits become unsuccessful after fine-tuning. We further find that fine-tuning only the edited layers is sufficient to effectively remove edits, while incurring only modest degradation in downstream performance. Surprisingly, fine-tuning non-edited layers leads to greater edit decay than all-layer fine-tuning. Besides, our activation space analysis reveals that fine-tuning produces a larger and more coherent representational shift, both in magnitude and direction, than KE. Overall, our study underscores the necessity of evaluating KE within the broader LLM application pipeline.

cs.CL↗

arXiv2026

Shortcut Mitigation via Spurious-Positive Samples

Shortcut mitigation strategies commonly rely on training data annotations, group-balanced held-out data or the presence of all groups, i.e., all combinations of (spurious) attributes and classes, in the training data. However, these requirements are rarely met in practice. We instead propose a method for targeted model analysis to identify a small set of instances in which the model relies on spurious attributes. Using that set and following ``this feature should not be used for prediction'' reasoning, we identify highly relevant neurons in an intermediate layer and regularize their impact. This ensures that models learn to depend on informative features rather than being right for the wrong reasons, thereby improving robustness without requiring additional balanced held-out data or annotations.

cs.LG↗

arXiv2026

Out of Spuriousity: Improving Robustness to Spurious Correlations without Group Annotations

Machine learning models are known to learn spurious correlations, i.e., features having strong relations with class labels but no causal relation. Relying on those correlations leads to poor performance in the data groups without these correlations and poor generalization ability. To improve the robustness of machine learning models to spurious correlations, we propose an approach to extract a subnetwork from a fully trained network that does not rely on spurious correlations. The subnetwork is found by the assumption that data points with the same spurious attribute will be close to each other in the representation space when training with ERM, then we employ supervised contrastive loss in a novel way to force models to unlearn the spurious connections. The increase in the worst-group performance of our approach contributes to strengthening the hypothesis that there exists a subnetwork in a fully trained dense network that is responsible for using only invariant features in classification tasks, therefore erasing the influence of spurious features even in the setup of multi spurious attributes and no prior knowledge of attributes labels.

cs.LG↗

arXiv2026

XNNTab -- Interpretable Neural Networks for Tabular Data using Sparse Autoencoders

In data-driven applications relying on tabular data, where interpretability is key, machine learning models such as decision trees and linear regression are applied. Although neural networks can provide higher predictive performance, they are not used because of their blackbox nature. In this work, we present XNNTab, a neural architecture that combines the expressiveness of neural networks and interpretability. XNNTab first learns highly non-linear feature representations, which are decomposed into monosemantic features using a sparse autoencoder (SAE). These features are then assigned human-interpretable concepts, making the overall model prediction intrinsically interpretable. XNNTab outperforms interpretable predictive models, and achieves comparable performance to its non-interpretable counterparts.

cs.LG↗

arXiv2026

Towards Interpretable Deep Neural Networks for Tabular Data

Tabular data is the foundation of many applications in fields such as finance and healthcare. Although DNNs tailored for tabular data achieve competitive predictive performance, they are blackboxes with little interpretability. We introduce XNNTab, a neural architecture that uses a sparse autoencoder (SAE) to learn a dictionary of monosemantic features within the latent space used for prediction. Using an automated method, we assign human-interpretable semantics to these features. This allows us to represent predictions as linear combinations of semantically meaningful components. Empirical evaluations demonstrate that XNNTab attains performance on par with or exceeding that of state-of-the-art, black-box neural models and classical machine learning approaches while being fully interpretable.

cs.LG↗

arXiv2026

This looks like what? Challenges and Future Research Directions for Part-Prototype Models

The growing interest in eXplainable Artificial Intelligence (XAI) has stimulated research on models with built-in interpretability, among which part-prototype models are particularly prominent. Part-Prototype Models (PPMs) classify inputs by comparing them to learned prototypes and provide human-understandable explanations of the form "this looks like that". Despite this intrinsic interpretability, PPMs have not yet emerged as a competitive alternative to post-hoc explanation methods. This survey reviews work published between 2019 and 2025 and derives a taxonomy of the challenges faced by current PPMs. The analysis reveals a diverse set of open problems. The main issue concerns the quality and number of learned prototypes. Further challenges include limited generalization across tasks and contexts, as well as methodological shortcomings such as non-standardized evaluation. Five broad research directions are identified: improving predictive performance, developing theoretically grounded architectures, establishing frameworks for human-AI collaboration, aligning models with human concepts, and defining robust metrics and benchmarks for evaluation. The survey aims to stimulate further research and promote intrinsically interpretable models for practical applications. A curated list of the surveyed papers is available at https://github.com/aix-group/ppm-survey.

cs.LG↗

arXiv2026

Tracing and Reversing Edits in LLMs

Knowledge editing methods (KEs) are a cost-effective way to update the factual content of large language models (LLMs), but they pose a dual-use risk. While KEs are beneficial for updating outdated or incorrect information, they can be exploited maliciously to implant misinformation or bias. In order to defend against these types of malicious manipulation, we need robust techniques that can reliably detect, interpret, and mitigate malicious edits. To that end, we introduce the tasks of tracing and reversing edits. We propose a novel method to infer the edited object entity, solely based on the modified weights, without access to the editing prompt or any other semantically similar prompts, with up to 99% accuracy. Further, we propose an effective and training-free method for reversing edits. Our method reverses up to 94% of the edits, and helps regain the original model's output distribution without access to any information about the edit. This method can further be repurposed to distinguish between edited and unedited weights. Our findings highlight the feasibility of tracing and reversing edits based on the edited weights, opening a new research direction for safeguarding LLMs against adversarial manipulations.

cs.CL↗

arXiv2026

Persuasion Tokens for Editing Factual Knowledge in LLMs

In-context knowledge editing (IKE) is a promising technique for updating Large Language Models (LLMs) with new information. However, IKE relies on lengthy, fact-specific demonstrations which are costly to create and consume significant context window space. In this paper, we introduce persuasion tokens (P-Tokens) -- special tokens trained to replicate the effect of IKE demonstrations, enabling efficient knowledge editing without requiring fact-specific demonstrations. We evaluate P-Tokens across two editing datasets and three LLMs, demonstrating performance comparable to, and often exceeding, IKE. We further find that editing performance is robust to distractors with small negative effects to neighboring facts, and that increasing the number of P-Tokens improves performance. Our work addresses key limitations of IKE and provides a more practical and scalable alternative for editing LLMs.

cs.CL↗

arXiv2025

Guiding LLMs to Generate High-Fidelity and High-Quality Counterfactual Explanations for Text Classification

The need for interpretability in deep learning has driven interest in counterfactual explanations, which identify minimal changes to an instance that change a model's prediction. Current counterfactual (CF) generation methods require task-specific fine-tuning and produce low-quality text. Large Language Models (LLMs), though effective for high-quality text generation, struggle with label-flipping counterfactuals (i.e., counterfactuals that change the prediction) without fine-tuning. We introduce two simple classifier-guided approaches to support counterfactual generation by LLMs, eliminating the need for fine-tuning while preserving the strengths of LLMs. Despite their simplicity, our methods outperform state-of-the-art counterfactual generation methods and are effective across different LLMs, highlighting the benefits of guiding counterfactual generation by LLMs with classifier information. We further show that data augmentation by our generated CFs can improve a classifier's robustness. Our analysis reveals a critical issue in counterfactual generation by LLMs: LLMs rely on parametric knowledge rather than faithfully following the classifier.

cs.CL↗

arXiv2025

Marcel: A Lightweight and Open-Source Conversational Agent for University Student Support

We present Marcel, a lightweight and open-source conversational agent designed to support prospective students with admission-related inquiries. The system aims to provide fast and personalized responses, while reducing workload of university staff. We employ retrieval-augmented generation to ground answers in university resources and to provide users with verifiable, contextually relevant information. We introduce a Frequently Asked Question (FAQ) retriever that maps user questions to knowledge-base entries, which allows administrators to steer retrieval, and improves over standard dense/hybrid retrieval strategies. The system is engineered for easy deployment in resource-constrained academic settings. We detail the system architecture, provide a technical evaluation of its components, and report insights from a real-world deployment.

cs.CL↗

arXiv2025

The Impact of Annotator Personas on LLM Behavior Across the Perspectivism Spectrum

In this work, we explore the capability of Large Language Models (LLMs) to annotate hate speech and abusiveness while considering predefined annotator personas within the strong-to-weak data perspectivism spectra. We evaluated LLM-generated annotations against existing annotator modeling techniques for perspective modeling. Our findings show that LLMs selectively use demographic attributes from the personas. We identified prototypical annotators, with persona features that show varying degrees of alignment with the original human annotators. Within the data perspectivism paradigm, annotator modeling techniques that do not explicitly rely on annotator information performed better under weak data perspectivism compared to both strong data perspectivism and human annotations, suggesting LLM-generated views tend towards aggregation despite subjective prompting. However, for more personalized datasets tailored to strong perspectivism, the performance of LLM annotator modeling approached, but did not exceed, human annotators.

cs.CL↗

arXiv2025

Position: Editing Large Language Models Poses Serious Safety Risks

Large Language Models (LLMs) contain large amounts of facts about the world. These facts can become outdated over time, which has led to the development of knowledge editing methods (KEs) that can change specific facts in LLMs with limited side effects. This position paper argues that editing LLMs poses serious safety risks that have been largely overlooked. First, we note the fact that KEs are widely available, computationally inexpensive, highly performant, and stealthy makes them an attractive tool for malicious actors. Second, we discuss malicious use cases of KEs, showing how KEs can be easily adapted for a variety of malicious purposes. Third, we highlight vulnerabilities in the AI ecosystem that allow unrestricted uploading and downloading of updated models without verification. Fourth, we argue that a lack of social and institutional awareness exacerbates this risk, and discuss the implications for different stakeholders. We call on the community to (i) research tamper-resistant models and countermeasures against malicious model editing, and (ii) actively engage in securing the AI ecosystem.

cs.CL↗

arXiv2025

Behavioral Analysis of Information Salience in Large Language Models

Large Language Models (LLMs) excel at text summarization, a task that requires models to select content based on its importance. However, the exact notion of salience that LLMs have internalized remains unclear. To bridge this gap, we introduce an explainable framework to systematically derive and investigate information salience in LLMs through their summarization behavior. Using length-controlled summarization as a behavioral probe into the content selection process, and tracing the answerability of Questions Under Discussion throughout, we derive a proxy for how models prioritize information. Our experiments on 13 models across four datasets reveal that LLMs have a nuanced, hierarchical notion of salience, generally consistent across model families and sizes. While models show highly consistent behavior and hence salience patterns, this notion of salience cannot be accessed through introspection, and only weakly correlates with human perceptions of information salience.

cs.CL↗

arXiv2025

Explanation format does not matter; but explanations do -- An Eggsbert study on explaining Bayesian Optimisation tasks

Bayesian Optimisation (BO) is a family of methods for finding optimal parameters when the underlying function to be optimised is unknown. BO is used, for example, for hyperparameter tuning in machine learning and as an expert support tool for tuning cyberphysical systems. For settings where humans are involved in the tuning task, methods have been developed to explain BO (Explainable Bayesian Optimization, XBO). However, there is little guidance on how to present XBO results to humans so that they can tune the system effectively and efficiently. In this paper, we investigate how the XBO explanation format affects users' task performance, task load, understanding and trust in XBO. We chose a task that is accessible to a wide range of users. Specifically, we set up an egg cooking scenario with 6 parameters that participants had to adjust to achieve a perfect soft-boiled egg. We compared three different explanation formats: a bar chart, a list of rules and a textual explanation in a between-subjects online study with 213 participants. Our results show that adding any type of explanation increases task success, reduces the number of trials needed to achieve success, and improves comprehension and confidence. While explanations add more information for participants to process, we found no increase in user task load. We also found that the aforementioned results were independent of the explanation format; all formats had a similar effect. This is an interesting finding for practical applications, as it suggests that explanations can be added to BO tuning tasks without the burden of designing or selecting specific explanation formats. In the future, it would be interesting to investigate scenarios of prolonged use of the explanation formats and whether they have different effects on users' mental models of the underlying system.

cs.HC↗

arXiv2025

An XAI-based Analysis of Shortcut Learning in Neural Networks

Machine learning models tend to learn spurious features - features that strongly correlate with target labels but are not causal. Existing approaches to mitigate models' dependence on spurious features work in some cases, but fail in others. In this paper, we systematically analyze how and where neural networks encode spurious correlations. We introduce the neuron spurious score, an XAI-based diagnostic measure to quantify a neuron's dependence on spurious features. We analyze both convolutional neural networks (CNNs) and vision transformers (ViTs) using architecture-specific methods. Our results show that spurious features are partially disentangled, but the degree of disentanglement varies across model architectures. Furthermore, we find that the assumptions behind existing mitigation methods are incomplete. Our results lay the groundwork for the development of novel methods to mitigate spurious correlations and make AI models safer to use in practice.

cs.LG↗

arXiv2025

Invariant Learning with Annotation-free Environments

Invariant learning is a promising approach to improve domain generalization compared to Empirical Risk Minimization (ERM). However, most invariant learning methods rely on the assumption that training examples are pre-partitioned into different known environments. We instead infer environments without the need for additional annotations, motivated by observations of the properties within the representation space of a trained ERM model. We show the preliminary effectiveness of our approach on the ColoredMNIST benchmark, achieving performance comparable to methods requiring explicit environment labels and on par with an annotation-free method that poses strong restrictions on the ERM reference model.

cs.LG↗

arXiv2025

How to Make LLMs Forget: On Reversing In-Context Knowledge Edits

In-context knowledge editing (IKE) enables efficient modification of large language model (LLM) outputs without parameter changes and at zero-cost. However, it can be misused to manipulate responses opaquely, e.g., insert misinformation or offensive content. Such malicious interventions could be incorporated into high-level wrapped APIs where the final input prompt is not shown to end-users. To address this issue, we investigate the detection and reversal of IKE-edits. First, we demonstrate that IKE-edits can be detected with high accuracy (F1 > 80\%) using only the top-10 output probabilities of the next token, even in a black-box setting, e.g. proprietary LLMs with limited output information. Further, we introduce the novel task of reversing IKE-edits using specially tuned reversal tokens. We explore using both continuous and discrete reversal tokens, achieving over 80\% accuracy in recovering original, unedited outputs across multiple LLMs. Our continuous reversal tokens prove particularly effective, with minimal impact on unedited prompts. Through analysis of output distributions, attention patterns, and token rankings, we provide insights into IKE's effects on LLMs and how reversal tokens mitigate them. This work represents a significant step towards enhancing LLM resilience against potential misuse of in-context editing, improving their transparency and trustworthiness.

cs.CL↗