Searcharxiv⌕ Search

arXiv subjects

Michael Alexander Riegler

Publications and source records attributed to Michael Alexander Riegler.

7 recordsLinked to original sources

Locating and Repairing Domain Shift in VLM Trajectory Planning

Vision-Language Models (VLMs) are increasingly adopted as end-to-end driving planners, yet they suffer from cross-city domain shift. Existing methods treat this shift as a single quantity addressed by uniform alignment or single-site editing. We challenge this view on two counts: (i) domain shift in VLM trajectory planners is inherently structured across modality and layer, and (ii) a locate-then-repair recipe matched to this structure outperforms every uniform-objective baseline. We analyse domain shift via activation patching and uncover two regularities. (F1) Modality axis. Image and text tokens exhibit distinct shift patterns in each VLM. (F2) Layer axis. The layer-wise shift partition is architecture-dependent and dataset-stable. These findings recast domain shift as a two-dimensional tensor indexed by modality (image vs. text tokens) and layer, motivating a simple principle: locate where shift occurs, and repair only there. We instantiate this principle in MoLaRx (Modality-Layer Repair), a locate-then-repair framework. The Causal Importance Score (CIS) estimates the domain shift as a modality-layer tensor and partitions the layers into functional zones. CIS-Guided Selective Repair (CISR) then applies a matched operator (Maximum Mean Discrepancy (MMD) / low-rank adaptation (LoRA) / freezing) to repair each zone, with the composition adapting automatically to each architecture. We evaluate MoLaRx on three architecturally distinct VLMs (LFM2-VL-1.6B, Qwen2-VL-2B, and PaliGemma2-3B) across two cross-city benchmarks (nuScenes and Argoverse 2). MoLaRx attains the lowest cross-city trajectory-error gap among adaptation methods in all six main-table settings, while using 50-60% fewer trainable parameters and 8.4-10.9 ms lower inference latency. In contrast, uniform MMD-only adaptation collapses in all six settings, highlighting the importance of structure-guided repair.

cs.CV↗

Restricting the Model, Missing the System: Measurement and Accountability in Offensive AI Governance

We show that the instruments used to measure AI offensive capability fail in two ways: (i) they overstate harm, and (ii) they credit the model with capability that belongs to the surrounding system. We argue that restricting access to a model is therefore necessary but not sufficient and that policy and procurement also need system-level, harm-grounded capability assessment. In June 2026, two frontier models were suspended under US export controls, reportedly prompted by a jailbreak that asked a model to read a codebase and fix its flaws. This finding measured an elicitation \emph{system} of model, prompt, and task. We support our argument with a study of an open-source framework in which lightweight large language model (LLM) agents coordinate through shared memory and evolutionary optimization, providing two pieces of evidence. First, jailbreak metrics overstate harm: over 225 swarm-generated attacks per target, LLM-as-judge scoring rated Claude Sonnet 4 compromised in 40% of attacks, yet manual verification found actionable harmful content in none, against a 45.8\% Effective Harm Rate for GPT-4o. Second, scaffolded evaluations misattribute capability: on a planted-vulnerability target, a full pipeline built around a 1.2B-parameter model recovers 9 of 9 weaknesses, while the same model without the hand-crafted components recovers 0 of 9 by crash verification and 2 of 9 by cited source line. Offensive capability is a property of model, scaffold, and evaluation protocol together. The duty to assess it should lie with whoever controls the system, shapes its behaviour, and can foresee what it will do. In our setting, that is the party who builds the harness around the model, a role that current regulation does not clearly cover.

cs.CR↗

Evaluation Cards: An Interpretive Layer for AI Evaluation Reporting

AI evaluation results are produced at scale but reported inconsistently across leaderboards, model cards, benchmark papers, and company blogs. The cost is interpretive: readers cannot reliably compare results across sources, identify what a report omits, or trace an aggregate claim to its underlying evidence. Recent efforts address isolated components but leave three gaps: they cover only narrow slices of the evaluation lifecycle and do not compose into a single interpretable record; they specify static representations that do not differentiate the questions different stakeholders bring to the same evidence; and they remain proposals on paper, lacking the extraction infrastructure required for adoption at scale. We present \EvalCards{}, an operational reporting layer that composes benchmark metadata, evaluation run data, and model metadata into a unified record. We (1) derive a reporting schema from a structured review of 52 papers and 10 stakeholder interviews, (2) implement four interpretive signals (reproducibility, documentation completeness, provenance and risk, and score comparability), rendered through reader modes calibrated to research and non-research audiences, and (3) deploy a monitoring tool that applies \EvalCards{} across 5,816 models, 635 benchmarks, and 101,843 results, surfacing systematic gaps in current reporting practice.

cs.AI↗

Who Evaluates AI's Social Impacts? Mapping Coverage and Gaps in First and Third Party Evaluations

Foundation models are increasingly central to high-stakes AI systems, and governance frameworks now depend on evaluations to assess their risks and capabilities. Although general capability evaluations are widespread, social impact assessments covering bias, fairness, privacy, environmental costs, and labor remain uneven. To characterize this landscape, we conduct the first comprehensive analysis of social impact evaluation reporting, examining 186 first-party release reports and 248 third-party evaluation sources, supplemented by developer interviews. We find a stark division of labor: first-party reporting is sparse, often superficial, and declining in areas like environmental impact and bias, while third-party evaluators provide broader, more rigorous coverage of bias, harmful content, and performance disparities. However, only developers can authoritatively report on data provenance, content moderation labor, costs, and infrastructure, yet interviews reveal these disclosures are deprioritized unless tied to product adoption or compliance. Current practices leave major gaps in assessing societal impacts, underscoring the need for policies that mandate developer transparency, strengthen independent evaluation ecosystems, and create shared infrastructure for aggregating third-party evaluations.

cs.CY↗

Classifying Dry Eye Disease Patients from Healthy Controls Using Machine Learning and Metabolomics Data

Dry eye disease is a common disorder of the ocular surface, leading patients to seek eye care. Clinical signs and symptoms are currently used to diagnose dry eye disease. Metabolomics, a method for analyzing biological systems, has been found helpful in identifying distinct metabolites in patients and in detecting metabolic profiles that may indicate dry eye disease at early stages. In this study, we explored using machine learning and metabolomics information to identify which cataract patients suffered from dry eye disease. As there is no one-size-fits-all machine learning model for metabolomics data, choosing the most suitable model can significantly affect the quality of predictions and subsequent metabolomics analyses. To address this challenge, we conducted a comparative analysis of nine machine learning models on three metabolomics data sets from cataract patients with and without dry eye disease. The models were evaluated and optimized using nested k-fold cross-validation. To assess the performance of these models, we selected a set of suitable evaluation metrics tailored to the data set's challenges. The logistic regression model overall performed the best, achieving the highest area under the curve score of 0.8378, balanced accuracy of 0.735, Matthew's correlation coefficient of 0.5147, an F1-score of 0.8513, and a specificity of 0.5667. Additionally, following the logistic regression, the XGBoost and Random Forest models also demonstrated good performance.

cs.CV↗

Principal Component Analysis based frameworks for efficient missing data imputation algorithms

Missing data is a commonly occurring problem in practice. Many imputation methods have been developed to fill in the missing entries. However, not all of them can scale to high-dimensional data, especially the multiple imputation techniques. Meanwhile, the data nowadays tends toward high-dimensional. Therefore, in this work, we propose Principal Component Analysis Imputation (PCAI), a simple but versatile framework based on Principal Component Analysis (PCA) to speed up the imputation process and alleviate memory issues of many available imputation techniques, without sacrificing the imputation quality in term of MSE. In addition, the frameworks can be used even when some or all of the missing features are categorical, or when the number of missing features is large. Next, we introduce PCA Imputation - Classification (PIC), an application of PCAI for classification problems with some adjustments. We validate our approach by experiments on various scenarios, which shows that PCAI and PIC can work with various imputation algorithms, including the state-of-the-art ones and improve the imputation speed significantly, while achieving competitive mean square error/classification accuracy compared to direct imputation (i.e., impute directly on the missing data).

cs.LG↗

MMSys'21 Grand Challenge on Detecting Cheapfakes

Cheapfake is a recently coined term that encompasses non-AI ("cheap") manipulations of multimedia content. Cheapfakes are known to be more prevalent than deepfakes. Cheapfake media can be created using editing software for image/video manipulations, or even without using any software, by simply altering the context of an image/video by sharing the media alongside misleading claims. This alteration of context is referred to as out-of-context (OOC) misuse} of media. OOC media is much harder to detect than fake media, since the images and videos are not tampered. In this challenge, we focus on detecting OOC images, and more specifically the misuse of real photographs with conflicting image captions in news items. The aim of this challenge is to develop and benchmark models that can be used to detect whether given samples (news image and associated captions) are OOC, based on the recently compiled COSMOS dataset.

cs.MM↗