Skip to main content
Search

Case Western Reserve University

10 related articles

Subscribe to updates
  1. arXiv

    LeanGuard blocks forbidden tool calls before they run: direct prompt-injection success on DTap falls from 57.9% to 5.9%

    Tool-using language-model agents often execute actions their domain policies forbid, and LeanGuard places policies written in Lean 4 between the model and its tools so that only admitted calls execute. On τ 2-bench, unguarded runs of five Qwen models executed at least 33 calls the Lean policy forbids, while LeanGuard executed none; on DTap, direct prompt-injection success fell from 57.9% to 5.9% and indirect success from 22.3% to 3.8%. Denials with feedback did not significantly change task success, and the read-only deployment blocks many benign writes.

  2. Nature Medicine

    A medical vision-language model that scores 0.895 on diagnosis and points back to the lesion

    Medical imaging foundation models have traded generality against pixel-level precision, and most return a label without evidence. EVLF-FM feeds both disease-level and pixel-level visual features into a language model, first with supervised fine-tuning and then with reinforcement learning on verifiable rewards, spanning six specialties and 10 imaging modalities. On internal diagnostic validation it reached the highest mean accuracy among evaluated models, 0.895, and averaged 0.842 Acc@0.5 for medical visual grounding across nine modalities.

  3. Diffractograms are not photos: an ImageNet-only model reads phase fractions from background noise, while physics pre-training locks onto Bragg rings

    Synchrotron X-ray diffraction produces a 2D diffractogram per second, and image models seem able to read α and β phase fractions of Ti–6Al–4V from them. Two models reach similar accuracy after fine-tuning on 250 experimental diffractograms, yet robustness tests and attention maps show the ImageNet-only model infers fractions from background and pixel noise, while the model pre-trained on simulated diffractograms relies on Bragg reflections. Accuracy alone does not validate a model.

  4. aclanthology.org

    Nine pretrained brain models, one prompt: NS-Copilot picks and runs them, reaching 57.16% macro F1 on three-class Alzheimer's EEG

    Pretrained neural-recording models differ in architecture and input format, so labs configure each one by hand before they can compare them. NS-Copilot puts nine EEG and spike models behind one natural-language interface, with planner, coder, controller and interpreter agents that select models, generate code and close the loop. Over 8 trials each on Alzheimer's, Parkinson's and working-memory benchmarks, its primary metrics were 57.16% macro F1, 71.32% balanced accuracy and 74.43%, above the listed baselines.

  5. arXiv

    Latent-MOPD lets one student absorb both the predictions and hidden states of multiple specialists, beating single-channel baselines on all nine math, code, and logic benchmarks

    The work introduces Latent-MOPD, a multi-teacher on-policy distillation method for LLMs that, to the authors' knowledge, is the first to integrate specialists at the representation level: it selects late-layer hidden-state targets according to the teacher-student relationship, bridges unequal hidden widths with a shared projection, groups updates by domain, and gradually shifts each teacher's supervision from hidden states to token predictions; in the same-family setting it outperforms token-only, representation-only, and uniform-averaging baselines on all nine benchmarks across math, code, and logic, surpasses the per-benchmark best teacher on a majority of them at equal parameter count, and with larger cross-family teachers it outperforms both single-channel baselines on all benchmarks.

  6. medRxiv

    Validating machine learning models in an independent external cohort: an ensemble model predicted incident hydronephrosis in spina bifida patients with a concordance index of 0.77 and classified bladder dysfunction with 71% accuracy

    This study adapted three previously developed machine learning models (a deep learning convolutional neural network using pressure-volume data, a deep learning imaging model using fluoroscopic imaging data, and an ensemble model averaging the risk data from both) to an independent external cohort of spina bifida patients who underwent videourodynamics at a single institution between 2016 and 2025, for prediction of incident hydronephrosis and classification of bladder dysfunction severity; among 70 patients in the hydronephrosis cohort, 13 (18%) developed incident hydronephrosis, the ensemble model achieved a concordance index of 0.77 (pressure-volume only: 0.70, imaging only: 0.72) and an overall AUROC of 0.75 (pressure-volume: 0.67, imaging: 0.

  7. A PCA-plus-local-differential-privacy framework for agricultural data sharing trades about 5.33% average accuracy loss for privacy on real datasets

    The work proposes a privacy-preserving data-sharing and collaboration framework for digital agriculture that combines Principal Component Analysis (PCA) dimensionality reduction with Laplacian-noise-based local differential privacy (ε-LDP), aggregates farmer data in a sandbox environment, and uses K-Means clustering and nearest-neighbor algorithms to recommend potential collaborators, enabling researchers to train personalized models via federated learning or directly on aggregated privacy-protected data; validated on the Wisconsin Farmer's Market and Crop Recommendation real-world datasets, model accuracy shows an average loss of about 5.33% compared with centralized raw-data training, with robustness against attacks such as membership inference assessed via power analysis.

  8. arXiv

    KANResDiff combines KAN spline time encoding with a local Schrödinger bridge for residual diffusion, cutting GED by up to 18.8% and raising HM-IoU32 by up to 7.7% on LIDC and ISIC3

    The work proposes KANResDiff, which learns local residual diffusion via Kolmogorov-Arnold Networks for ambiguous medical image segmentation: it replaces MLP linear time embeddings with B-spline-based Independent Time Encoding to strengthen independence across inference stages, and injects a deterministic residual prior with learnable weights through a Residual Schrödinger Bridge, achieving state-of-the-art GED and HM-IoU on the two public datasets LIDC and ISIC3 with maximum improvements of 16.8% and 7.7% respectively while keeping competitive MDM performance.

  9. medRxiv

    Privacy-Aware Distillation of Large Language Models for Enhanced Multimorbidity Scoring

    This study introduces and evaluates a privacy-preserving knowledge distillation framework in which CTGAN-generated synthetic cohorts matching UK Biobank distributions are used to elicit multimorbidity scores from three teacher LLMs (GPT-4o, Gemini, DeepSeek) under zero-shot prompting, and compact student models (CoLLMs) are then trained to mimic those scores, enabling application to real UK Biobank data (N = 439,221) for multimorbidity scoring without exposing patient-level data to third-party APIs, with evaluation against the Charlson (CCI) and Elixhauser (ECI) indices via survival analysis, genome-wide association studies, and polygenic risk score associations.

  10. Artificial intelligence in trigeminal neuralgia: trigeminal nerve segmentation and neurovascular conflict detection: a systematic review and meta-analysis

    This systematic review and meta-analysis pooled 5 studies covering 577 patients with confirmed trigeminal neuralgia to evaluate deep learning and machine learning models for automated trigeminal nerve segmentation and neurovascular conflict detection on MRI, finding a pooled sensitivity of 71% (95% CI: 65-76%), specificity of 92% (95% CI: 88-94%), and AUC of 0.91 (95% CI: 0.88-0.93), suggesting high diagnostic accuracy for AI models in this task.

You’re all caught up