MetaDent: Labeling Clinical Images for Vision-Language Models in Dentistry

Vision-Language Models (VLMs) have demonstrated significant potential in medical image analysis, yet their application in intraoral photography remains largely underexplored due to the lack of fine-grained, annotated datasets and comprehensive benchmarks. To address this, we present MetaDent, a comprehensive resource that includes (1) a novel and large-scale dentistry image dataset collected from clinical, public, and web sources; (2) a semi-structured annotation framework designed to capture the hierarchical and clinically nuanced nature of dental photography; and (3) comprehensive benchmark suites for evaluating state-of-the-art VLMs on clinical image understanding. Our labeling approach combines a high-level image summary with point-by-point, free-text descriptions of abnormalities. This method enables rich, scalable, and task-agnostic representations. We curated 60,669 dental images from diverse sources and annotated a representative subset of 2,588 images using this meta-labeling scheme. Leveraging Large Language Models (LLMs), we derive standardized benchmarks: approximately 15K Visual Question Answering (VQA) pairs and an 18-class multi-label classification dataset, which we validated with human review and error analysis to justify that the LLM-driven transition reliably preserves fidelity and semantic accuracy. We then evaluate state-of-the-art VLMs across VQA, classification, and image captioning tasks. Quantitative results reveal that even the most advanced models struggle with a fine-grained understanding of intraoral scenes, achieving moderate accuracy and producing inconsistent or incomplete descriptions in image captioning. We publicly release our dataset, annotations, and tools to foster reproducible research and accelerate the development of vision-language systems for dental applications.

Human vs. Machine Deception: Distinguishing AI-Generated and Human-Written Fake News Using Ensemble Learning

The rapid adoption of large language models has introduced a new class of AI-generated fake news that coexists with traditional human-written misinformation, raising important questions about how these two forms of deceptive content differ and how reliably they can be distinguished. This study examines linguistic, structural, and emotional differences between human-written and AI-generated fake news and evaluates machine learning and ensemble-based methods for distinguishing these content types. A document-level feature representation is constructed using sentence structure, lexical diversity, punctuation patterns, readability indices, and emotion-based features capturing affective dimensions such as fear, anger, joy, sadness, trust, and anticipation. Multiple classification models, including logistic regression, random forest, support vector machines, extreme gradient boosting, and a neural network, are applied alongside an ensemble framework that aggregates predictions across models. Model performance is assessed using accuracy and area under the receiver operating characteristic curve. The results show strong and consistent classification performance, with readability-based features emerging as the most informative predictors and AI-generated text exhibiting more uniform stylistic patterns. Ensemble learning provides modest but consistent improvements over individual models. These findings indicate that stylistic and structural properties of text provide a robust basis for distinguishing AI-generated misinformation from human-written fake news.

BITS Pilani at SemEval-2026 Task 9: Structured Supervised Fine-Tuning with DPO Refinement for Polarization Detection

The POLAR SemEval-2026 Shared Task aims to detect online polarization and focuses on the classification and identification of multilingual, multicultural, and multi-event polarization. Accurate computational detection of online polarization is challenging due to nuanced rhetoric, implicit framing, and the high cost of human-in-the-loop annotation. Building on recent findings that contextual prompting enables large language models to function as strong polarization detectors, we present a two-stage approach for detecting political polarization in social media text that combines structured supervised fine-tuning with Direct Preference Optimization (DPO) refinement. We fine-tune Qwen 2.5-7B-Instruct with LoRA using an interpretable slot-filling template (target, claim type, manifestation checklist, and justification). We then apply DPO with automatically generated preference pairs to reduce costly false negatives. Experiments on the SemEval 2026 POLAR shared task dataset show that preference-based refinement improves both accuracy and decreases false negatives without extra annotation. On the English development set, DPO increases recall from 0.5085 to 0.7797 and improves macro-F1 by ~5 points.

Representation geometry shapes task performance in vision-language modeling for CT enterography

Computed tomography (CT) enterography is a primary imaging modality for assessing inflammatory bowel disease (IBD), yet the representational choices that best support automated analysis of this modality are unknown. We present the first study of vision-language transfer learning on abdominal CT enterography and identify two main findings. First, mean pooling of slice embeddings gives better categorical disease assessment (59.2\% three-class accuracy), whereas attention pooling gives better cross-modal retrieval (0.235 text-to-image MRR). This pattern holds across all LoRA configurations tested and suggests that the two aggregators emphasize different properties of the learned representation. Second, per-slice tissue contrast matters more than broader spatial coverage: multi-window RGB encoding, which maps complementary Hounsfield Unit windows to RGB channels, outperforms all strategies that increase spatial coverage through multiplanar sampling, and in this setting adding coronal and sagittal views reduces classification performance. For report generation, fine-tuning without retrieval context yields within-1 severity accuracy at the prevalence-matched chance level (70.4\% vs.\ 71\% random), suggesting little learned ordering beyond the class distribution. Retrieval-augmented generation (RAG) improves this across all configurations, scoring 7--14 percentage points above the chance baseline and improving ordinal MAE from 0.98 to 0.80--0.89. A three-teacher pseudolabel framework enables all comparisons without expert annotations. Together, these findings provide the first baselines for this underexplored modality and offer practical guidance for building vision-language systems for volumetric medical imaging.

EmergentBridge: Improving Zero-Shot Cross-Modal Transfer in Unified Multimodal Embedding Models

Unified multimodal embedding spaces underpin practical applications such as cross-modal retrieval and zero-shot recognition. In many real deployments, however, supervision is available only for a small subset of modality pairs (e.g., image--text), leaving \emph{unpaired} modality pairs (e.g., audio$\leftrightarrow$depth, infrared$\leftrightarrow$audio) weakly connected and thus performing poorly on zero-shot transfer. Addressing this sparse-pairing regime is therefore essential for scaling unified embedding systems to new tasks without curating exhaustive pairwise data. We propose \textbf{EmergentBridge}, an embedding-level bridging framework that improves performance on these unpaired pairs \emph{without requiring exhaustive pairwise supervision}. Our key observation is that naively aligning a new modality to a synthesized proxy embedding can introduce \emph{gradient interference}, degrading the anchor-alignment structure that existing retrieval/classification relies on. EmergentBridge addresses this by (i) learning a mapping that produces a \emph{noisy bridge anchor} (a proxy embedding of an already-aligned modality) from an anchor embedding, and (ii) enforcing proxy alignment only in the subspace orthogonal to the anchor-alignment direction, preserving anchor alignment while strengthening non-anchor connectivity. Across nine datasets spanning multiple modalities, EmergentBridge consistently outperforms prior binding baselines on zero-shot classification and retrieval, demonstrating strong emergent alignment.

Is There Knowledge Left to Extract? Evidence of Fragility in Medically Fine-Tuned Vision-Language Models

Vision-language models (VLMs) are increasingly adapted through domain-specific fine-tuning, yet it remains unclear whether this improves reasoning beyond superficial visual cues, particularly in high-stakes domains like medicine. We evaluate four paired open-source VLMs (LLaVA vs. LLaVA-Med; Gemma vs. MedGemma) across four medical imaging tasks of increasing difficulty: brain tumor, pneumonia, skin cancer, and histopathology classification. We find that performance degrades toward near-random levels as task difficulty increases, indicating limited clinical reasoning. Medical fine-tuning provides no consistent advantage, and models are highly sensitive to prompt formulation, with minor changes causing large swings in accuracy and refusal rates. To test whether closed-form VQA suppresses latent knowledge, we introduce a description-based pipeline where models generate image descriptions that a text-only model (GPT-5.1) uses for diagnosis. This recovers a limited additional signal but remains bounded by task difficulty. Analysis of vision encoder embeddings further shows that failures stem from both weak visual representations and downstream reasoning. Overall, medical VLM performance is fragile, prompt-dependent, and not reliably improved by domain-specific fine-tuning.

BMdataset: A Musicologically Curated LilyPond Dataset

Symbolic music research has relied almost exclusively on MIDI-based datasets; text-based engraving formats such as LilyPond remain unexplored for music understanding. We present BMdataset, a musicologically curated dataset of 393 LilyPond scores (2,646 movements) transcribed by experts directly from original Baroque manuscripts, with metadata covering composer, musical form, instrumentation, and sectional attributes. Building on this resource, we introduce LilyBERT (weights can be found at https://huggingface.co/csc-unipd/lilybert), a CodeBERT-based encoder adapted to symbolic music through vocabulary extension with 115 LilyPond-specific tokens and masked language model pre-training. Linear probing on the out-of-domain Mutopia corpus shows that, despite its modest size (~90M tokens), fine-tuning on BMdataset alone outperforms continuous pre-training on the full PDMX corpus (~15B tokens) for both composer and style classification, demonstrating that small, expertly curated datasets can be more effective than large, noisy corpora for music understanding. Combining broad pre-training with domain-specific fine-tuning yields the best results overall (84.3% composer accuracy), confirming that the two data regimes are complementary. We release the dataset, tokenizer, and model to establish a baseline for representation learning on LilyPond.

LPLCv2: An Expanded Dataset for Fine-Grained License Plate Legibility Classification

Modern Automatic License Plate Recognition (ALPR) systems achieve outstanding performance in controlled, well-defined scenarios. However, large-scale real-world usage remains challenging due to low-quality imaging devices, compression artifacts, and suboptimal camera installation. Identifying illegible license plates (LPs) has recently become feasible through a dedicated benchmark; however, its impact has been limited by its small size and annotation errors. In this work, we expand the original benchmark to over three times the size with two extra capture days, revise its annotations and introduce novel labels. LP-level annotations include bounding boxes, text, and legibility level, while vehicle-level annotations comprise make, model, type, and color. Image-level annotations feature camera identity, capture conditions (e.g., rain and faulty cameras), acquisition time, and day ID. We present a novel training procedure featuring an Exponential Moving Average-based loss function and a refined learning rate scheduler, addressing common mistakes in testing. These improvements enable a baseline model to achieve an 89.5% F1-score on the test set, considerably surpassing the previous state of the art. We further introduce a novel protocol to explicitly addresses camera contamination between training and evaluation splits, where results show a small impact. Dataset and code are publicly available at https://github.com/lmlwojcik/LPLCv2-Dataset.

ProbeLogits: Kernel-Level LLM Inference Primitives for AI-Native Operating Systems

An OS kernel that runs LLM inference internally can read logit distributions before any text is generated -- and act on them as a governance primitive. I present ProbeLogits, a kernel-level operation that performs a single forward pass and reads specific token logits to classify agent actions as safe or dangerous, with zero learned parameters. On a 260-prompt OS action benchmark (9 categories including adversarial attacks), ProbeLogits achieves F1=0.980, Precision=1.000, and Recall=0.960 using a general-purpose 7B model at 4-bit quantization. On ToxicChat (1,000 human-annotated real conversations), it achieves F1=0.790 at default calibration strength $α$=1.0, improving to F1=0.837 at $α$=0.5 -- 89% of Llama Guard 3's F1~0.939 with zero learned parameters. A key design contribution is the calibration strength $α$, which serves as a deployment-time policy knob rather than a learned hyperparameter. By adjusting $α$, the OS can enforce strict policies for privileged operations ($α\geq 0.8$, maximizing recall) or relaxed policies for conversational agents ($α$=0.5, maximizing precision). Contextual calibration improves accuracy from 64.8% to 97.3% on the custom benchmark. I implement ProbeLogits within Anima OS, a bare-metal x86_64 OS written in 80,400 lines of Rust. Because agent actions must pass through kernel-mediated host functions, ProbeLogits enforcement operates below the WASM sandbox boundary, making it significantly harder to circumvent than application-layer classifiers. Each classification costs 65ms on 7B -- fast enough for per-action governance. I also show that treating KV cache as process state enables checkpoint, restore, and fork operations analogous to traditional process management. To my knowledge, no prior system exposes LLM logit vectors as OS-level governance primitives.

BadSkill: Backdoor Attacks on Agent Skills via Model-in-Skill Poisoning

Agent ecosystems increasingly rely on installable skills to extend functionality, and some skills bundle learned model artifacts as part of their execution logic. This creates a supply-chain risk that is not captured by prompt injection or ordinary plugin misuse: a third-party skill may appear benign while concealing malicious behavior inside its bundled model. We present BadSkill, a backdoor attack formulation that targets this model-in-skill threat surface. In BadSkill, an adversary publishes a seemingly benign skill whose embedded model is backdoor-fine-tuned to activate a hidden payload only when routine skill parameters satisfy attacker-chosen semantic trigger combinations. To realize this attack, we train the embedded classifier with a composite objective that combines classification loss, margin-based separation, and poison-focused optimization, and evaluate it in an OpenClaw-inspired simulation environment that preserves third-party skill installation and execution while enabling controlled multi-model study. Our benchmark spans 13 skills, including 8 triggered tasks and 5 non-trigger control skills, with a combined main evaluation set of 571 negative-class queries and 396 trigger-aligned queries. Across eight architectures (494M--7.1B parameters) from five model families, BadSkill achieves up to 99.5\% average attack success rate (ASR) across the eight triggered skills while maintaining strong benign-side accuracy on negative-class queries. In poison-rate sweeps on the standard test split, a 3\% poison rate already yields 91.7\% ASR. The attack remains effective across the evaluated model scales and under five text perturbation types. These findings identify model-bearing skills as a distinct model supply-chain risk in agent ecosystems and motivate stronger provenance verification and behavioral vetting for third-party skill artifacts.