Skip to main content
Search

RWTH Aachen University

9 related articles

Subscribe to updates
  1. arXiv

    Constrained optimization stops sampling constraints uniformly: keeping recently violated ones cuts iterate error by two orders of magnitude on synthetic problems

    For stochastic-gradient optimization with very many inequality constraints, an adaptive constraint-sampling method stops drawing each step's constraint uniformly from all constraints and instead maintains an "important set" of recently violated constraints, sampling from that set or from all constraints with fixed probabilities. On 100-dimensional synthetic convex problems it lowers the final Euclidean distance of the iterates to the constrained minimizer by about two orders of magnitude on average for ABSGD, Projection-SGD and Huber-penalty SGD, and by one to two orders for APDSG; in Dubins Car control-barrier-function synthesis it lowers maximum constraint violation by half an order to one order of magnitude.

  2. bioRxiv

    Every base of the human genome swapped in silico: about 3.12 billion positions form a directed dependency network with seven recurring architectures

    Using the DNA language model Nucleotide Transformer v3 on the JUPITER supercomputer, the team substituted each of roughly 3.12 billion positions of the telomere-to-telomere human reference genome with each of the three alternative bases and recorded how model predictions changed elsewhere, yielding a genome-wide directed map of nucleotide dependencies. These dependencies preferentially link evolutionarily conserved positions and repeatedly assemble into seven geometric structures called Relons, some outside current annotation. Combined with CADD, dependency features raised intronic-variant pathogenicity AUPRC from 0.299 to 0.356.

  3. medRxiv

    Therapy sessions can now run in silico: raising therapist empathy prompts lifts every motivational-interviewing integrity score

    Psychotherapy research is constrained by costly trials and clinical features that are hard to manipulate at random. A preregistered proof-of-concept study paired therapist and patient agents built on a locally deployed Llama-3.3-70B-Instruct to generate 512 automated motivational interviewing sessions, manipulating only the therapist prompts for empathy, affirmation and alliance building. Among 480 sessions retained for analysis, all four motivational interviewing treatment integrity global ratings were higher under high common factors, with adjusted mean differences of 0.54 to 0.95 points.

  4. arXiv

    Post-training slices a language model's belief state into a causal quotient: reward-null information is mostly rerouted or rescaled, and only sustained weight decay erases it

    In hidden-Markov worlds where pretraining recovers Bayesian belief states, the authors define an exact reward-null kernel from a reward that reads only a coarse function of the hidden state, and separately measure whether reward-null information stays linearly decodable, whether decisions causally depend on it, and how much activation variance it occupies; post-training mostly reroutes or rescales this information while leaving it decodable, a KL anchor keeps decisions using it, unanchored objectives let decisions stop using it, and only prolonged weight decay erases distinctions that neither reward nor next-token prediction can see, with open language models showing the same dissociation in in-context belief geometry.

  5. bioRxiv

    TRIDENT-2 predicts chemical toxicity across Eukaryota from 560,780 assays with average median absolute error of 1.76 to 3.80

    The authors present TRIDENT-2, a multimodal artificial intelligence model trained on 560,780 toxicity assays spanning 82,775 chemicals, 6,793 species, and multiple exposure scenarios to predict chemical toxicity across evolutionarily diverse eukaryotic species, reporting an average median absolute error of 1.76 to 3.80 and remaining accurate across broad chemical and taxonomic distances, which allows toxicity assessment for species and chemicals beyond current experimental evidence.

  6. arXiv

    SnapLog turns kitchen videos into event logs via ViT frame embeddings, K-Means segmentation and R(2+1)D few-shot classification, reaching 90% and 85% Top-3

    The work presents SnapLog, a pipeline that encodes each video frame with a pre-trained ViT, performs temporal segmentation through a frame-wise similarity matrix with K-Means and a greedy merge, then applies generalized few-shot classification with a 34-layer R(2+1)D encoder pre-trained on Sports-1M plus a linear head to convert raw video into timestamped event logs (either deterministic or uncertainty-aware logs that retain label probability distributions); on 16 brownie-baking videos from TUM Kitchen and 24 kitchen-cleaning videos from Epic Kitchens-100 it achieves frame-wise segmentation accuracy of 74.3% and 67.7%, and with augmentation plus 100 overlapping clips Top-3 classification accuracy of 90.1% and 85.

  7. arXiv

    DP-RGMI splits differential privacy's performance loss on 594,000 chest X-rays into encoder geometry and task-head utilization

    The authors introduce DP-RGMI, a framework that treats differentially private training as a structured transformation of representation space and decomposes performance degradation into representation displacement, spectral effective dimension, and a utilization gap defined as the difference between linear-probe and end-to-end AUROC; across more than 594,000 chest X-rays from four datasets and three pretrained initializations (ImageNet, DINOv3, MIMIC-CXR), with PadChest as the primary dataset (110,525 frontal images, 22,045 test images), they find that strong privacy consistently leaves linear separability largely preserved while a utilization gap persists (G = 8.0 for ImageNet at ε = 1.0, 3.4 for MIMIC at ε = 0.7, 6.1 for DINOv3 at ε = 0.

  8. Benchmarking AI Vibe Coding for Clinical Statistical Analysis: A Structured Evaluation in Pulmonary Hypertension Research

    This study had five current large language models (GPT 5.3, Claude Sonnet 4.6, Gemini 2.5 Flash, Perplexity, and Grok) reproduce three statistical tasks from a published clinical workflow under identical datasets and standardised prompts—descriptive table generation, Kaplan–Meier survival analysis, and Cox proportional hazards modelling—and compared their outputs with analyses by two experts trained in mathematical statistics on grouping correctness, numerical accuracy, missing-value handling, and quality of generated R code, finding that all five models produced correct descriptive statistics once dataset variables were specified explicitly, that two models failed the initial descriptive benchmark because of variable-name ambiguity but recovered after prompt clarification, that all models

  9. Clinical Code Mapping with LLM Tool Use: A Pilot for Automated Data Extraction of Medication and Diagnosis Information from Unstructured Clinical Notes

    This pilot study anonymized 35 German doctor's notes from five patients, built one pipeline for medication extraction and mapping and two for diagnoses (one RAG-based and one agentic AI), ran them with three open-weight LLMs on a local GPU-PC, and found that medication name extraction reached an F1 of 0.95 and medication mapping 0.78, while diagnosis coding did not exceed an F1 of 0.12 and broad-category mapping reached 0.18, leading the authors to conclude that LLMs are suitable for medication information extraction for research databases but that current state-of-the-art open-weight models are not accurate enough for a clinical setting where patient treatment would depend on LLM performance.

You’re all caught up