In many forms of reasoning, including arithmetic reasoning, generalizing across superficial changes in input format is effortless for humans: anyone who can solve 2+5 can also solve 'two plus five'. In contrast, LLMs are more brittle to surface variations of the prompts: for example, they solve numeric arithmetic problems almost perfectly but are substantially less accurate on verbal renditions of the same problems. Here, we ask whether generalization across formats can be predicted from the models' internals. Using attribution patching, we first independently localize the circuit that each model recruits to solve numeric arithmetic problems (2+5) vs. verbal ones, in three languages: English ('two plus five'), Spanish ('dos m\'as cinco'), and Italian ('due pi\`u cinque'); then, we test whether overlap with the model's own numeric circuit predicts its generalization to the verbal formats. Indeed, we find support for this idea at three levels: circuit overlap accounts for the relative difficulty of the three verbal formats, for which models generalize best, and for which items are solved correctly, rivaling supervised probes while requiring no labeled data.
Andrea Gregor de Varda, Sana Pandey, Pengrui Han et al.· 0 citations
Foundation models are beginning to reshape brain-signal analysis by moving the field beyond task-specific decoding pipelines toward reusable models pretrained on broad neural datasets. Magnetoencephalography (MEG) is a compelling but still underdeveloped target for this shift: it captures human cortical dynamics at millisecond resolution while offering stronger spatial interpretability than EEG, making it especially valuable for source-resolved studies of perception, language, cognition, and clinical brain function. Yet MEG foundation models remain at an early stage, with only a small number of MEG-specific and MEG-inclusive multi-modal models, modest pretraining corpora, and emerging but still limited benchmarks. This perspective lays down the basic concepts needed to understand MEG foundation models and provides a didactic overview of the field's key design choices, including tokenization, sensor- versus source-space representations, sensor-geometry encoding, backbone architectures, self-supervised objectives, and pretraining data. We then offer a roadmap for future development, organized around native MEG pretraining, adaptation of EEG foundation models, transfer from generic time-series models, and multi-modal integration with EEG, fMRI, MRI, behaviour, and stimulus features. We highlight the need for coordinated infrastructure, including diverse and reusable MEG datasets, rigorous evaluation across subjects, sites, tasks, and clinical settings, and responsible data-sharing practices that address consent, privacy, access, and governance.
Philipp Th\"olke, Hamza Abdelhedi, Yorguin Mantilla-Ramos et al.· 0 citations
Expert pruning reduces the memory and serving cost of Mixture-of-Experts (MoE) models by removing low-importance experts identified by the router, assuming router probabilities provide a reliable importance signal. We observe that this assumption breaks down under over-dispersed routing, a regime associated with aggressive load-balancing during training, in which tokens are distributed nearly uniformly across experts and importance signals collapse. In this regime, perplexity does not predict downstream task accuracy: on gpt-oss-20B, the lowest-perplexity pruning configuration yields the worst mathematical reasoning, while the highest-perplexity configuration preserves it. This does not occur under standard routing (e.g., Mixtral-8x7B-Instruct), where perplexity and accuracy degrade together. Pruning under over-dispersed routing also exposes a capability trade-off in which no single scoring metric dominates: activation-aware scoring preserves mathematical reasoning but severely degrades knowledge-intensive science (an 18-point gap on GPQA), whereas frequency-based scoring exhibits the reverse. We propose Minimax Expert Score Allocation (MESA), a domain-aware method that iteratively boosts importance scores for experts serving whichever domain is currently worst-affected, minimizing worst-case domain degradation rather than average accuracy. At 25% expert pruning MESA achieves the smallest worst-case degradation across domains, outperforming activation-aware baselines on 7 of 11 benchmarks at a correspondingly reduced memory footprint, and it generalizes to gpt-oss-120B, Gemma-4-26B-A4B, and OLMoE-1B-7B. Our results indicate that over-dispersed routing is a qualitatively distinct pruning regime in which standard assumptions fail, and that recognizing it is a prerequisite for principled expert pruning of load-balanced MoE models.
Berkcan Kapusuzoglu, Connor Pryor, Sangwoo Cho et al.· 0 citations
Large language models deployed in high-stakes settings frequently generate plausible but ungrounded claims. Standard retrieval-augmented generation (RAG) pipelines offer limited remedy, since they retrieve isolated passages without tracking cross-document evidence relationships or quantifying uncertainty. We introduce GRACE (Graph-grounded Reflective Agent Copilot Engine), a framework that deconstructs LLM responses into atomic claims and grounds them against trusted knowledge priors within a weighted bipartite graph. Edge weights encode the closeness of each claim to the priors, enabling weighted centrality analysis that classifies claims as Grounded, Refuted, or Boundary. Such classification identifies not just hallucinations but also novel or contested claims at the frontier of the model's knowledge. To efficiently allocate human or agent resources, we formulate a Return on Attention (RoA) objective that defers a claim to expert review only when its priority-weighted uncertainty exceeds the cost of verification. Claims verified by experts are promoted to new evidence anchors, closing a validator-LLM evolutionary loop that expands the knowledge base across iterations. We evaluate GRACE across multiple language models and on datasets spanning both general and domain-specific knowledge. Our results show that our knowledge base serves as a reliable foundation for retrieval that outperforms RAG baselines, and that the RoA framework efficiently selects valuable boundary knowledge for expert verification. These findings demonstrate that graph-structured representations combined with expert-in-the-loop verification can mitigate hallucination at the system level rather than at the generation level. Code available at https://github.com/johnsk95/grace_code
John Seon Keun Yi, Joshua R. Minot, Dokyun Lee· 0 citations
Reach audiences
Advertise in front of researchers, engineers, and readers.
Learning rich medical concept representations is essential for EHR prediction. Text-attributed knowledge graphs (TKGs) provide a natural foundation by organizing heterogeneous medical relations together with textual semantics. However, most existing encoders process concepts uniformly across patients, despite the fact that a code's meaning and predictive value depend on patient-specific clinical context and trajectory. Learning patient-personalized concept representations from TKGs introduces two key challenges: (1) deciding how much KG context to incorporate for each observed code, and (2) aligning semantic information with the patient-specific relational structure. We propose REFINE, a KG-aware budgeted LLM graph refinement framework for patient-personalized medical concept encoding. Starting from a global TKG, REFINE constructs patient-specific temporal graphs. A sequential reinforcement learning policy selects a personalized KG expansion budget for each observed code. The resulting patient graph is processed by a heterogeneous GNN to capture relation-aware structural dependencies, while a frozen LLM uses graph-aware soft prompts to semantically refine concept representations. Experiments on MIMIC-III and MIMIC-IV show that REFINE consistently improves diverse EHR backbones, outperforms strong baselines, and demonstrates robust gains across component ablation, KG selection, and data insufficiency.
Mohsen Nayebi Kerdabadi, Arya Hadizadeh Moghaddam, Dongjie Wang et al.· 0 citations
Multilingual language models often produce inconsistent answers to semantically equivalent questions across languages, motivating methods to improve cross-lingual consistency (CLC). However, existing methods are typically evaluated using different models, tasks, and protocols, leaving their relative strengths unclear. In this work, we present a unified evaluation of representative CLC-enhancement methods for question answering, spanning inference-time interventions and post-training approaches across three model families and three closed-form benchmarks. The results show that post-training methods are generally more reliable, with direct distribution alignment consistently improving CLC across all model-dataset combinations, while other methods are more sensitive to answer format and the breadth of language coverage. Notably, cross-domain transfer is limited unless source and target tasks share similar output formats. We further investigate whether CLC enhancement hurts models' ability to respond differently *when needed*, that is, when asked culture-dependent questions. Across two benchmarks of culturally diverse question answering, we find no systematic degradation in controlled closed-form evaluation, whereas open-ended generation reveals occasional accuracy reductions, particularly for non-English responses. Our work highlights the need to evaluate CLC enhancement for both cross-domain robustness and culturally appropriate variation, informing future work in post-training and benchmark development.
Jirui Qi, Mingyang Wang, Hinrich Sch\"utze et al.· 0 citations
Chinese online comments often convey social meaning through indirect and playful language that is hard to interpret without context. Existing evaluations largely organize items around predefined phenomena or controlled pragmatic categories, leaving open whether models can distinguish plausible readings of what a naturally occurring comment is doing in a particular exchange. We introduce a benchmark for evaluating whether LLMs can recover such situated pragmatic meanings. From more than 200,000 public Chinese social media interaction records, we construct 4,735 human-validated diagnostic items, each pairing a target comment with reconstructed preceding context and plausible misreadings. We evaluate eight LLMs as both question writers and solvers in a cross-writer setting. The task is challenging: the strongest model achieves 81.42% leave-writer-out accuracy. Across all eight models, the mean leave-writer-out accuracy is 68.70% while human accuracy was 90.8%. Case analysis shows that models often recognize broad irony or playfulness while misidentifying the mechanism or interactional move.
Shiwei Hong, Junjie Ma, Emma Jiren Wang et al.· 0 citations
Real-world dynamics are inherently compositional: multiple entities move simultaneously within a shared scene, each exhibiting distinct motion patterns. Yet most existing video representations encode motion globally, without explicitly capturing localized motion for individual entities. Crucially, motion is defined relative to a global reference frame, including camera motion and scene layout. However, localized embeddings are often computed from cropped images or obtained by masking features after encoding, discarding the context needed to interpret motion. To address this, we introduce a promptable localized motion representation that produces persistent embeddings for user-specified regions defined by spatial masks. Rather than cropping the input or masking features, our model processes the full video and conditions motion encoding directly on the queried region. This yields temporally consistent, region-addressable embeddings that isolate local dynamics while retaining the global context required for disambiguation. We demonstrate object-level motion transfer, enabling controlled composition of dynamic scenes. Beyond generative control, our embeddings support localized action classification in multi-actor videos. Across both tasks, our approach improves controllability and outperforms global representations localized through cropping or post-hoc masking. Project Page: https://compvis.github.io/WhatMoves
Frank Fundel, Malek Ben Alaya, Thomas Ressler-Antal et al.· 0 citations
Recognizing specific objects onboarded without a labeled training set recurs across manufacturing and service robotics, yet the conventional renderable prior, a computer-aided-design (CAD) model, is often unavailable. Two-dimensional capture supplies no shape prior, and frozen foundation features fail on geometrically similar, low-texture industrial parts. We ask what a short object-centric scan buys for recognition beyond the captured images themselves: each object is reconstructed with 3D Gaussian Splatting (3DGS), summarized into a per-class shape prototype, and fused with frozen DINOv2 image features. First, the scan recovers the recognition value of CAD without CAD: geometry from RGB-D depth (on T-LESS), 3DGS, and CAD gives comparable recognition (tied on HOPE, within 1.6 points on T-LESS); 3DGS is only a convenient route to a point cloud. Second, the payoff is governed by how recognizable the shape is: on shape-distinctive household objects (HOPE) geometry alone reaches 0.920 versus image-only 0.832, a ceiling below which fixed-weight fusion (0.872) sits. On shape-confusable textureless industrial parts (T-LESS) the gain is modest but consistent (0.560 to 0.591 fused, above both single signals). Third, the prior is complementary, not uniformly additive: it rescues far more image failures than it breaks successes, and its benefit grows under partial occlusion. Finally, the worth lies in geometry, not rendered pixels: 3DGS renderings do not help the image side, and frozen-feature recognition is nearly lighting-invariant (within 2.5 points). The study is scoped to recognition, not the BOP pose benchmark.
Decompensation represents a critical transition in the course of cirrhosis, yet clinicians have limited non-invasive tools to reliably predict its onset. In this study, we propose a novel imaging-based approach that leverages large-scale computer vision models to analyze routine abdominal ultrasound images and extract predictive features beyond those captured by traditional laboratory-based risk scores. Ultrasound is widely available, low cost, and suitable for longitudinal surveillance, making it an attractive modality for scalable risk stratification and long-term follow-up. Our framework integrates automated ultrasound data processing with modern deep learning architectures to identify patients at high risk of decompensation prior to the occurrence of clinical deterioration. This non-invasive strategy offers a practical complement to existing clinical scoring systems and may enable earlier, more proactive management of patients with compensated cirrhosis.
Guangyi Zhang, Peiyun Ni, Eugene Cheah et al.· 0 citations
Purpose: Increased number of chest radiograph (CXR) scans create a triage bottleneck, queueing urgent examinations behind routine ones. Existing AI tools are predominantly unimodal binary classifiers lacking severity awareness, and multimodal systems are rarely benchmarked against expert radiologists. To this end, we developed a multimodal deep learning framework for joint severity triage, pathology detection, and native visual explanation. Approach: We propose the cross-modal triage network (CMTN), fusing a Swin Transformer V2 visual encoder with a PubMedBERT text encoder via gated cross-attention. The CMTN was trained on 34,639 image-text pairs (12,489 patients) from MIMIC-CXR-JPG, optimizing an ordinal focal loss for four-tier severity triage and binary cross-entropy for 14 pathologies. Beyond quantitative benchmarking, attention heatmaps were evaluated against a blinded expert radiologist in a two-phase clinical audit comparing model triage output to expert severity assessment (100 cases) and grading spatial-semantic concordance (116 heatmaps). Results: The CMTN achieved strong ordinal agreement with reference labels (quadratic weighted kappa [QWK] = 0.9341, 95\% CI: 0.9219 to 0.9449) and macro-AUROC of 0.9970 across 14 pathologies, with 34~ms latency, outperforming the state-of-the-art BioViL multimodal baseline (QWK = 0.7679). However, the blinded Phase I clinical audit revealed substantially lower agreement with genuine radiologist judgment (QWK = 0.1399). Phase II found 54.3\% of heatmaps achieved clinically acceptable spatial localization. Conclusions: The CMTN demonstrated an efficient multimodal architecture for CXR triage. The divergence between algorithmic and radiologist agreement demonstrates that benchmark performance against NLP-derived labels is insufficient, highlighting the need for radiologist-labeled ground truth before clinical deployment.
This paper analyzes the implementation of blockchain-based integrity mechanisms in Greek Fiscal Electronic Mechanisms (FEMs) and the central tax information system eSEND. The study examines the cryptographic architecture of fiscal devices, including Electronic Cash Registers, Fiscal Printers, Fiscal Signing Machines, and FEMAS devices, which implement double or triple hash-chain structures to ensure transaction immutability. The transmission protocol between fiscal devices and the central database is also evaluated with respect to encryption, sequential validation, and blockchain verification. In contrast, the architecture of Electronic Invoicing Provider Services and the myDATA central platform is analyzed, highlighting the absence of blockchain-based integrity guarantees. The comparison demonstrates that hardware-based fiscal mechanisms provide stronger guarantees for transaction completeness and tamper resistance than purely software-based invoicing infrastructures. The findings highlight architectural weaknesses in the current e-invoicing framework and propose improvements for ensuring transaction integrity in digital tax ecosystems.
