Skip to main content
Search

University of Amsterdam

9 related articles

Subscribe to updates
  1. arXiv

    A few rewritten decision rules in a third-party skill make agents answer wrong while running normally, and most scanners miss it

    Third-party skills in public registries can be exploited by a poisoning method that edits only decision rules while leaving the task and interface intact. Without real user requests or ground-truth answers, StealthyPoison uses synthetic requests and clean executions by a surrogate agent as references to iteratively rewrite formulas, thresholds and branches. Across 566 eligible skills, 5,244 requests and more than 26,000 registry entries, the frozen poisoned skills reached micro-averaged attack success of 31.0%–31.9% on three held-out models, close to the surrogate's 32.9%.

  2. arXiv

    Same object, new camera angle, wrong answer: LieVLM predicts local rotations from geometry tokens and lifts ScanRefer grounding accuracy by 6.9 points

    Multi-view 3D vision-language models suffer viewpoint-induced semantic drift: the same object and query produce weaker, diffuse or shifted target responses when the camera moves. LieVLM has geometry tokens predict local rotation parameters and channel-wise gates, applies length-preserving rotations to projected visual features within learned three-dimensional subspaces, and gates the identity-relative correction back into the original representation. On ScanRefer, LieVLM-8B reaches 64.5 Acc@0.25 and 57.2 Acc@0.5, 6.9 and 6.3 points above the published VG LLM-8B under the same proposal-refinement protocol.

  3. arXiv

    Local LLM inference is not private: an EM probe beside a Raspberry Pi recovers generated tokens and prompts

    On a Raspberry Pi 5 running LLaMA-3.2-1B-Instruct locally, electromagnetic leakage captured by a passive probe can identify generated tokens and semantically reconstruct user prompts, showing that local execution alone does not keep prompts and outputs confidential.

  4. arXiv

    SDECast uses neural SDEs for continuous-time probabilistic weather forecasting, recovering drift dynamics in a simulated geophysical flow and producing skillful global hourly forecasts up to five days

    The authors introduce SDECast, a Neural Stochastic Differential Equation (SDE) framework for continuous-time probabilistic weather forecasting that extends SDE Matching to learn stochastic dynamics directly in physical space without repeated SDE simulation during training; on a simulated geophysical flow, SDECast recovers meaningful drift dynamics and faithfully reproduces the underlying continuous-time behavior, and it then scales to global weather forecasting at hourly resolution, producing skillful probabilistic forecasts for lead times up to five days.

  5. arXiv

    LiFT lets looped DiTs keep improving beyond training depth: L/2 and XL/2 beat larger dense DiTs with fewer parameters and less inference compute

    LiFT supervises each loop of a recurrent DiT toward a point on a straight path from the model's initial velocity estimate to the flow-matching target, so a trained checkpoint can run far more inference loops than its training depth; on class-conditional ImageNet, L/2 and XL/2 checkpoints that reallocate inference compute toward recurrence reach lower FID than larger dense DiTs with fewer parameters and less inference compute, while at B/2 the dense model remains stronger.

  6. arXiv

    CARD pairs cluster-level LoRA with decoding-time preference vectors, taking 10 of 12 metric settings across six LaMP and LongLaMP tasks

    CARD introduces a hierarchical framework for personalized text generation: it clusters users by shared stylistic patterns and trains group-specific LoRA adapters, derives lightweight user preference vectors through implicit preference learning that contrasts user-authored text with cluster-level generations, and injects personalization at inference only via low-rank logit corrections, ranking first in 10 of 12 settings across six LaMP and LongLaMP tasks and two metrics while remaining stable for low-resource users, across model scales, and in storage efficiency.

  7. arXiv

    TG-OT matches features on a topological cylinder via unbalanced optimal transport, achieving fully automatic segmentation-free CCTA-IVUS registration on 47 paired cases (Dicectl=0.99, Sc=0.96, DiceL=0.69)

    The work proposes TG-OT, a fully automatic CCTA-IVUS registration framework: lightweight CNNs first predict calcifications, bifurcations, and lumen radii on the topological (θ, z) cylinder (with the IVUS network additionally detecting guidewire artifacts), and the frozen detectors are then integrated directly into a differentiable registration pipeline that optimizes centerline warping parameters driven by an unbalanced Sinkhorn optimal transport loss on the cylindrical geometry plus a Dice term, complemented by a lumen radius matching term; on N=47 paired CCTA-IVUS cases from the IMPACT study at Erasmus University Medical Center in a 5-fold cross-validation setup, it reaches longitudinal Dicectl=0.99, rotational Sc=0.96, and lumen DiceL=0.

  8. medRxiv

    Artificial Intelligence-Enhanced Electrocardiography for Detection and Prediction of Hypertrophic Cardiomyopathy across Monogenic and Polygenic Susceptibility

    In 1,095 carriers of pathogenic or likely pathogenic sarcomere variants across three international centers, a previously validated AI-ECG model applied to 12-lead ECG images yielded an AUROC of 0.91 for the HCM phenotype at baseline and 0.92 for manifest HCM, a higher AI-ECG score among genotype-positive/phenotype-negative individuals predicted incident HCM during follow-up (unadjusted HR 1.55 per 1-SD; adjusted HR 1.38), and in 57,007 UK Biobank participants the AI-ECG score and a polygenic risk score were independent and additive, with adjusted odds of HCM of 60.2 when both were high.

  9. Real-World External Validation of Artificial Intelligence-Based Full-Vessel Segmentation for Intracoronary Optical Coherence Tomography

    This retrospective, single-center external validation study enrolled 100 consecutive patients undergoing clinically indicated OCT and used the previously developed OCT-AID algorithm to perform automated pixelwise full-vessel labeling of 2560 analyzable frames, comparing it frame by frame against an expert manual reference standard; agreement was excellent for calcified plaque identification (κ=0.88) and quantification (intraclass correlation coefficients 0.79–0.93), close to interobserver variability, reasonable for lipid plaque identification and quantification (κ=0.68; lipid arc intraclass correlation coefficient 0.79; minimum fibrous cap thickness intraclass correlation coefficient 0.

You’re all caught up