GUI grounding models report over 85% accuracy on standard benchmarks, yet drop 27-56 percentage points when instructions require spatial reasoning rather than direct element naming. Current benchmarks miss this because they evaluate each screenshot once with a single fixed instruction. We introduce GUI-Perturbed, a controlled perturbation framework that independently varies visual scenes and instructions to measure grounding robustness. Evaluating three 7B models from the same architecture lineage, we find that relational instructions cause systematic accuracy collapse across all models, a 70% browser zoom produces statistically significant degradation, and rank-8 LoRA fine-tuning with augmented data degrades performance rather than improving it. By perturbing along independent axes, GUI-Perturbed isolates which specific capability axes are affected-spatial reasoning, visual robustness, reasoning calibration-providing diagnostic signal that aggregate benchmarks cannot. We release the dataset, augmentation pipeline, and a fine-tuned model.

The rapid rise in AI conference submissions has driven increasing exploration of large language models (LLMs) for peer review support. However, LLM-based reviewers often generate superficial, formulaic comments lacking substantive, evidence-grounded feedback. We attribute this to the underutilization of two key components of human reviewing: explicit rubrics and contextual grounding in existing work. To address this, we introduce REVIEWBENCH, a benchmark evaluating review text according to paper-specific rubrics derived from official guidelines, the paper's content, and human-written reviews. We further propose REVIEWGROUNDER, a rubric-guided, tool-integrated multi-agent framework that decomposes reviewing into drafting and grounding stages, enriching shallow drafts via targeted evidence consolidation. Experiments on REVIEWBENCH show that REVIEWGROUNDER, using a Phi-4-14B-based drafter and a GPT-OSS-120B-based grounding stage, consistently outperforms baselines with substantially stronger/larger backbones (e.g., GPT-4.1 and DeepSeek-R1-670B) in both alignment with human judgments and rubric-based review quality across 8 dimensions. The code is available \href{https://github.com/EigenTom/ReviewGrounder}{here}.

The field of machine ethics aims to build Artificial Moral Agents (AMAs) to better understand morality and make AI agents safer. To do so, many approaches encode human moral intuition as a set of axioms on actions e.g., do not harm, you must help others. However, this introduces (at least) two limitations for future AMAs. First, it does not consider the agent's purposes in performing the action. Second, it assumes that we humans can enumerate our moral intuition. This paper explores formalizing a moral procedure that alleviates these two limitations. We specifically consider Kantian ethics and present a multi-sorted quantified modal logic we call the Formula of the Universal Law Logic (FULL). The FULL formalizes Kant's first formulation of the categorical imperative, the Formula of the Universal Law (FUL), and concepts such as causality and agency. We demonstrate on three cases from Kantian ethics that the FULL can reason to evaluate agents' actions for certain purposes without built-in moral intuition, given that it has sufficient (non-normative) background knowledge. Therefore, the FULL is a contribution towards more robust and autonomous AMAs, and a more formal understanding of Kantian ethics.

Monitoring LLM safety at scale requires balancing cost and accuracy: a cheap latent-space probe can screen every input, but hard cases should be escalated to a more expensive expert. Existing cascades delegate based on probe uncertainty, but uncertainty is a poor proxy for delegation benefit, as it ignores whether the expert would actually correct the error. To address this problem, we introduce Calibrate-Then-Delegate (CTD), a model-cascade approach that provides probabilistic guarantees on the computation cost while enabling instance-level (streaming) decisions. CTD builds on a novel delegation value (DV) probe, a lightweight model operating on the same internal representations as the safety probe that directly predicts the benefit of escalation. To enforce budget constraints, CTD calibrates a threshold on the DV signal using held-out data via multiple hypothesis testing, yielding finite-sample guarantees on the delegation rate. Evaluated on four safety datasets, CTD consistently outperforms uncertainty-based delegation at every budget level, avoids harmful over-delegation, and adapts budget allocation to input difficulty without requiring group labels.

We introduce Metric-Aware Principal Component Analysis (MAPCA), a unified framework for scale-invariant representation learning based on the generalised eigenproblem max Tr(W^T Sigma W) subject to W^T M W = I, where M is a symmetric positive definite metric matrix. The choice of M determines the representation geometry. The canonical beta-family M(beta) = Sigma^beta, beta in [0,1], provides continuous spectral bias control between standard PCA (beta=0) and output whitening (beta=1), with condition number kappa(beta) = (lambda_1/lambda_p)^(1-beta) decreasing monotonically to isotropy. The diagonal metric M = D = diag(Sigma) recovers Invariant PCA (IPCA), a method rooted in Frisch (1928) diagonal regression, as a distinct member of the broader framework. We prove that scale invariance holds if and only if the metric transforms as M_tilde = CMC under rescaling C, a condition satisfied exactly by IPCA but not by the general beta-family at intermediate values. Beyond its classical interpretation, MAPCA provides a geometric language that unifies several self-supervised learning objectives. Barlow Twins and ZCA whitening correspond to beta=1 (output whitening); VICReg's variance term corresponds to the diagonal metric. A key finding is that W-MSE, despite being described as a whitening-based method, corresponds to M = Sigma^{-1} (beta = -1), outside the spectral compression range entirely and in the opposite spectral direction to Barlow Twins. This distinction between input and output whitening is invisible at the level of loss functions and becomes precise only within the MAPCA framework.

Transistor topology optimization is a critical step in standard cell design, directly dictating diffusion sharing efficiency and downstream routability. However, identifying optimal topologies remains a persistent bottleneck, as conventional exhaustive search methods become computationally intractable with increasing circuit complexity in advanced nodes. This paper introduces TOPCELL, a novel and scalable framework that reformulates high-dimensional topology exploration as a generative task using Large Language Models (LLMs). We employ Group Relative Policy Optimization (GRPO) to fine-tune the model, aligning its topology optimization strategy with logical (circuit) and spatial (layout) constraints. Experimental results within an industrial flow targeting an advanced 2nm technology node demonstrate that TOPCELL significantly outperforms foundation models in discovering routable, physically-aware topologies. When integrated into a state-of-the-art (SOTA) automation flow for a 7nm library generation task, TOPCELL exhibits robust zero-shot generalization and matches the layout quality of exhaustive solvers while achieving an 85.91x speedup.

Fraud detection on graph data can be viewed as a demanding task that requires distinguishing between different types of nodes. Because graph neural networks (GNNs) are naturally suited for processing information encoded in graph form through their message-passing operations, methods based on GNN models have increasingly attracted attention in the fraud detection domain. However, fraud graphs inherently exhibit relation camouflage, high heterophily, and class imbalance, causing most GNNs to underperform in fraud detection tasks. To address these challenges, this paper proposes a Graph-Based Fraud Detection Model with Dual-Path Graph Filtering (DPF-GFD). DPF-GFD first applies a beta wavelet-based operator to the original graph to capture key structural patterns. It then constructs a similarity graph from distance-based node representations and applies an improved low-pass filter. The embeddings from the original and similarity graphs are fused through supervised representation learning to obtain node features, which are finally used by an ensemble tree model to assess the fraud risk of unlabeled nodes. Unlike existing single-graph smoothing approaches, DPF-GFD introduces a frequency-complementary dual-path filtering paradigm tailored for fraud detection, explicitly decoupling structural anomaly modeling and feature similarity modeling. This design enables more discriminative and stable node representations in highly heterophilous and imbalanced fraud graphs. Comprehensive experiments on four real-world financial fraud detection datasets demonstrate the effectiveness of our proposed method.

The Spatial-Temporal Graph Attention Network (ST-GAT) framework was created to serve as an explainable GNN-based solution for detecting bank distress early warning signs and for conducting macro-prudential surveillance of the interbank system in the United States. The ST-GAT framework models 8,103 FDIC insured institutions across 58 quarterly snapshots (2010Q1-2024Q2). Bilateral exposures were reconstructed from publicly available FDIC Call Reports using maximum entropy estimation to produce a dynamic directed weighted graph. The framework achieves the highest AUPRC among all GNN architectures (0.939 +/- 0.010), trailing only XGBoost (0.944). Ablation analysis confirms the BiLSTM temporal component contributes +0.020 AUPRC; temporal attention weights exhibit a monotonically decreasing pattern consistent with long-run structural vulnerability weighting. Permutation importance identifies ROA (0.309) and NPL Ratio (0.252) as dominant predictors, consistent with post-mortem analyses of the 2023 regional banking crisis. All data are publicly available FDIC Call Reports and FRED series; all code and results are released.

Financial crime costs U.S. institutions over $32 billion each year. Although AI tools for fraud detection have become more advanced, their use in real-world systems still faces a major obstacle: many of these models operate as black boxes that cannot provide the transparent, auditable explanations required by regulations such as OCC Bulletin 2011-12 and Federal Reserve SR 11-7. This study makes three main contributions. First, it offers a thorough evaluation of explanation quality across faithfulness (sufficiency and comprehensiveness at k=5, 10, and 15) and stability (Kendall's W across 30 bootstrap samples). XGBoost paired with TreeExplainer achieves near-perfect stability (W=0.9912), while LSTM with DeepExplainer shows weak results (W=0.4962). Second, the paper introduces the SHAP-Guided Adaptive Ensemble (SGAE), which dynamically adjusts per-transaction ensemble weights based on SHAP attribution agreement, achieving the highest AUC-ROC among all tested models (0.8837 held-out; 0.9245 cross-validation). Third, a complete three-architecture evaluation of LSTM, Transformer, and GNN-GraphSAGE on the full 590,540-transaction IEEE-CIS dataset is provided, with GNN-GraphSAGE achieving AUC-ROC 0.9248 and F1=0.6013. All results are mapped directly to OCC, SR 11-7, and BSA-AML regulatory compliance requirements.

Reliable evaluation of anomaly detection methods in multivariate time series remains an open challenge, largely due to the limitations of existing benchmark datasets. Current resources often lack fine-grained anomaly annotations, do not provide explicit intervariable and temporal dependencies, and offer little insight into the underlying generative mechanisms. These shortcomings hinder the development and rigorous comparison of detection models, especially those targeting interpretable and variable-specific outputs. To address this gap, we introduce Fun-TSG, a fully customizable time series generator designed to support high-quality evaluation of anomaly detection systems. Our tool enables both fully automated generation, based on randomly sampled dependency structures and anomaly types, and manual generation through user-defined equations and anomaly configurations. In both cases, it provides full transparency over the data generation process, including access to ground-truth anomaly labels at the variable and timestamp levels. Fun-TSG supports the creation of diverse, interpretable, and reproducible benchmarking scenarios, enabling fine-grained performance analysis for both classical and modern anomaly detection models.

Hate speech detection in Devanagari-scripted social media memes presents compounded challenges: multimodal content structure, script-specific linguistic complexity, and extreme data scarcity in low-resource settings. This paper presents our system for the CHiPSAL 2026 shared task, addressing both Subtask A (binary hate speech detection) and Subtask B (three-class sentiment classification: positive, neutral, negative). We propose a hybrid cross-modal attention fusion architecture that combines CLIP (ViT-B/32) for visual encoding with BGE-M3 for multilingual text representation, connected through 4-head self-attention and a learnable gating network that dynamically weights modality contributions on a per-sample basis. Systematic evaluation across eight model configurations demonstrates that explicit cross-modal reasoning achieves a 5.9% F1-macro improvement over text-only baselines on Subtask A, while uncovering two unexpected but critical findings: English-centric vision models exhibit near-random performance on Devanagari script, and standard ensemble methods catastrophically degrade under data scarcity (N nearly equal to 850 per fold) due to correlated overfitting. The code can be accessed at https://github.com/Tri-Yantra-Technologies/MEME-Fusion/

Large Language Models utilizing reasoning techniques improve task performance but incur significant latency and token costs due to verbose generation. Existing automatic prompt optimization(APO) frameworks target task accuracy exclusively at the expense of generating long reasoning traces. We propose Cost-Regularized Optimization of Prompts (CROP), an APO method that introduces regularization on response length by generating textual feedback in addition to standard accuracy feedback. This forces the optimization process to produce prompts that elicit concise responses containing only critical information and reasoning. We evaluate our approach on complex reasoning datasets, specifically GSM8K, LogiQA and BIG-Bench Hard. We achieved an 80.6\% reduction in token consumption while maintaining competitive accuracy, seeing only a nominal decline in performance. This presents a pragmatic solution for deploying token-efficient and cost-effective agentic AI systems in production pipelines.

A claim has been circulating on social media and practitioner forums that Chinese prompts are more token-efficient than English for LLM coding tasks, potentially reducing costs by up to 40\%. This claim has influenced developers to consider switching to Chinese for ``vibe coding'' to save on API costs. In this paper, we conduct a rigorous empirical study using SWE-bench Lite, a benchmark of software engineering tasks, to evaluate whether this claim of Chinese token efficiency holds up to scrutiny. Our results reveal three key findings: First, the efficiency advantage of Chinese is not observed. Second, token cost varies by model architecture in ways that defy simple assumptions: while MiniMax-2.7 shows 1.28x higher token costs for Chinese, GLM-5 actually consumes fewer tokens with Chinese prompts. Third, and most importantly, we found that the success rate when prompting in Chinese is generally lower than in English across all models we tested. We also measure cost efficiency as expected cost per successful task -- jointly accounting for token consumption and task resolution rate. These findings should be interpreted as preliminary evidence rather than a definitive conclusion, given the limited number of models evaluated and the narrow set of benchmarks tested due to resource constraints; they indicate that language effects on token cost are model-dependent, and that practitioners should not expect cost savings or performance gains just by switching their prompt language to Chinese.

As deep neural networks are deployed in safety-critical domains such as autonomous driving and medical diagnosis, stakeholders need explanations that are interpretable but also trustworthy with formal guarantees. Existing XAI methods fall short: heuristic attribution techniques (e.g., LIME, Integrated Gradients) highlight influential features but offer no mathematical guarantees about decision boundaries, while formal methods verify robustness yet remain untargeted, analyzing the nearest boundary regardless of whether it represents a critical risk. In safety-critical systems, not all misclassifications carry equal consequences; confusing a "Stop" sign for a "60 kph" sign is far more dangerous than confusing it with a "No Passing" sign. We introduce ViTaX (Verified and Targeted Explanations), a formal XAI framework that generates targeted semifactual explanations with mathematical guarantees. For a given input (class y) and a user-specified critical alternative (class t), ViTaX: (1) identifies the minimal feature subset most sensitive to the y->t transition, and (2) applies formal reachability analysis to guarantee that perturbing these features by epsilon cannot flip the classification to t. We formalize this through Targeted epsilon-Robustness, certifying whether a feature subset remains robust under perturbation toward a specific target class. ViTaX is the first method to provide formally guaranteed explanations of a model's resilience against user-identified alternatives. Evaluations on MNIST, GTSRB, EMNIST, and TaxiNet demonstrate over 30% fidelity improvement with minimal explanation cardinality.

This paper proposes a machine learning assisted portfolio optimization framework designed for low data environments and regime uncertainty. We construct a teacher student learning pipeline in which a Conditional Value at Risk (CVaR) optimizer generates supervisory labels, and neural models (Bayesian and deterministic) are trained using both real and synthetically augmented data. The synthetic data is generated using a factor based model with t copula residuals, enabling training beyond the limited real sample of 104 labeled observations. We evaluate four student models under a structured experimental framework comprising (i) controlled synthetic experiments (3 x 5 seed grid), (ii) in-distribution real market evaluation (C2A) and (iii) cross-universe generalization (D2A). In real-market settings, models are deployed using a rolling evaluation protocol where a frozen pretrained model is periodically fine tuned on recent observations and reset to its base state, ensuring stability while allowing limited adaptation. Results show that student models can match or outperform the CVaR teacher in several settings, while achieving improved robustness under regime shifts and reduced turnover. These findings suggest that hybrid optimization learning approaches can enhance portfolio construction in data constrained environments

Multimodal emotion recognition in conversations aims to infer utterance-level emotions by jointly modeling textual, acoustic, and visual cues within context. Despite recent progress, key challenges remain, including redundant cross-modal information, imperfect semantic alignment, and insufficient modeling of high-order speaker interactions. To address these issues, we propose a framework that combines dual-space feature disentanglement with dual-branch graph learning. A shared encoder and modality-specific encoders are used to separate modality-invariant and modality-specific representations. The invariant features are modeled by a Fourier graph neural network to capture global consistency and complementary patterns, with a frequency-domain contrastive objective to enhance discriminability. In parallel, a speaker-aware hypergraph is constructed over modality-specific features to model high-order interactions, along with a speaker-consistency constraint to maintain coherent semantics. Finally, the two branches are fused for utterance-level emotion prediction. Experiments on IEMOCAP and MELD demonstrate that the proposed method achieves superior performance over strong baselines, validating its effectiveness.

Domain reweighting can improve sample efficiency and downstream generalization, but data-mixture optimization for multimodal midtraining remains largely unexplored. Current multimodal training recipes tune mixtures along a single dimension, typically data format or task type. We introduce MixAtlas, a method that produces benchmark-targeted data recipes that can be inspected, adapted, and transferred to new corpora. MixAtlas decomposes the training corpus along two axes: image concepts (10 visual-domain clusters discovered via CLIP embeddings) and task supervision (5 objective types including captioning, OCR, grounding, detection, and VQA). Using small proxy models (Qwen2-0.5B) paired with a Gaussian-process surrogate and GP-UCB acquisition, MixAtlas searches the resulting mixture space with the same proxy budget as regression-based baselines but finds better-performing mixtures. We evaluate on 10 benchmarks spanning visual understanding, document reasoning, and multimodal reasoning. On Qwen2-7B, optimized mixtures improve average performance by 8.5%-17.6% over the strongest baseline; on Qwen2.5-7B, gains are 1.0%-3.3%. Both settings reach baseline-equivalent training loss in up to 2 times fewer steps. Recipes discovered on 0.5B proxies transfer to 7B-scale training across Qwen model families.

Large language model (LLM) performance depends heavily on prompt design, yet prompt construction is often described and applied inconsistently. Our purpose was to derive a reference framework for structuring LLM prompts. This paper presents PICCO, a framework derived through a rigorous synthesis of 11 previously published prompting frameworks identified through a multi-database search. The analysis yields two main contributions. First, it proposes a taxonomy that distinguishes prompt frameworks, prompt elements, prompt generation, prompting techniques, and prompt engineering as related but non-equivalent concepts. Second, it derives a five-element reference architecture for prompt generation: Persona, Instructions, Context, Constraints, and Output (PICCO). For each element, we define its function, scope, and relationship to other elements, with the goal of improving conceptual clarity and supporting more systematic prompt design. Finally, to support application of the framework, we outline key concepts relevant to implementation, including prompting techniques (e.g., zero-shot, few-shot, chain-of-thought, ensembling, decomposition, and self-critique, with selected variants), human and automated approaches to iterative prompt engineering, responsible prompting considerations such as security, privacy, bias, and trust, and priorities for future research. This work is a conceptual and methodological contribution: it formalizes a common structure for prompt specification and comparison, but does not claim empirical validation of PICCO as an optimization method.

State Space Models (SSMs) such as Mamba have become a popular alternative to Transformer models, due to their reduced memory consumption and higher throughput at generation compared to their Attention-based counterparts. On the other hand, the community has built up a considerable body of knowledge on how to train Transformers, and many pretrained Transformer models are readily available. To facilitate the adoption of SSMs while leveraging existing pretrained Transformers, we aim to identify an effective recipe to distill an Attention-based model into a Mamba-like architecture. In prior work on cross-architecture distillation, however, it has been shown that a naïve distillation procedure from Transformers to Mamba fails to preserve the original teacher performance, a limitation often overcome with hybrid solutions combining Attention and SSM blocks. The key argument from our work is that, by equipping Mamba with a principled initialization, we can recover an overall better recipe for cross-architectural distillation. To this end, we propose a principled two-stage approach: first, we distill knowledge from a traditional Transformer into a linearized version of Attention, using an adaptation of the kernel trick. Then, we distill the linearized version into an adapted Mamba model that does not use any Attention block. Overall, the distilled Mamba model is able to preserve the original Pythia-1B Transformer performance in downstream tasks, maintaining a perplexity of 14.11 close to the teacher's 13.86. To show the efficacy of our recipe, we conduct thorough ablations at 1B scale with 10B tokens varying sequence mixer architecture, scaling analysis on model sizes and total distillation tokens, and a sensitivity analysis on tokens allocation between stages.

We train a 318M-parameter Transformer language model from scratch on a curated corpus of 1.56 billion tokens of pure Classical Chinese, with zero English characters or Arabic numerals. Through systematic out-of-distribution (OOD) testing, we investigate whether the model can distinguish known from unknown inputs, and crucially, whether it can express this distinction in its generated text. We find a clear dissociation between internal and external uncertainty. Internally, the model exhibits a perplexity jump ratio of 2.39x between real and fabricated historical events (p = 8.9e-11, n = 92 per group), with semi-fabricated events (real figures + fictional events) showing the highest perplexity (4.24x, p = 1.1e-16), demonstrating genuine factual encoding beyond syntactic pattern matching. Externally, however, the model never learns to express uncertainty: classical Chinese epistemic markers appear at lower rates for OOD questions (3.5%) than for in-distribution questions (8.3%, p = 0.023), reflecting rhetorical conventions rather than genuine metacognition. We replicate both findings across three languages (Classical Chinese, English, Japanese), three writing systems, and eight models from 110M to 1.56B parameters. We further show that uncertainty expression frequency is determined entirely by training data conventions, with Classical Chinese models showing a "humility paradox" (more hedging for known topics), while Japanese models almost never hedge. We argue that metacognitive expression -- the ability to say "I don't know" -- does not emerge from language modeling alone and requires explicit training signals such as RLHF.

Rare diseases affect over 300 million people worldwide and are characterized by complex care pathways, limited clinical expertise, and substantial unmet communication needs throughout the long patient journey. Recent advances in large language models (LLMs) offer new opportunities to support patient education and communication, yet their application in rare diseases remains unclear. We conducted a scoping review of studies published between January 2022 and March 2026 across major databases, identifying 12 studies on LLM-based rare disease patient education and communication. Data were extracted on study characteristics, application scenarios, model usage, and evaluation methods, and synthesized using descriptive and qualitative analyses. The literature is highly recent and dominated by general-purpose models, particularly ChatGPT. Most studies focus on patient question answering using curated question sets, with limited use of real-world data or longitudinal communication scenarios. Evaluations are primarily centered on accuracy, with limited attention to patient-centered dimensions such as readability, empathy, and communication quality. Multilingual communication is rarely addressed. Overall, the field remains at an early stage. Future research should prioritize patient-centered design, domain-adapted methods, and real-world deployment to support safe, adaptive, and effective communication in rare diseases.

Large Language Model (LLM) agents have demonstrated remarkable capabilities in reasoning and tool use, yet they often suffer from rigid, reactive control flows that limit their adaptability and efficiency. Most existing frameworks rely on fixed pipelines or failure-triggered reflection, causing agents to act impulsively or correct errors only after they occur. In this paper, we introduce Heartbeat-Driven Autonomous Thinking Activity Scheduling, a mechanism that enables proactive, adaptive, and continuous self-regulation. Mirroring the natural rhythm of human cognition, our system employs a periodic ``heartbeat'' mechanism to orchestrate a dynamic repertoire of cognitive modules (e.g., Planner, Critic, Recaller, Dreamer). Unlike traditional approaches that rely on hard-coded symbolic rules or immediate reactive triggers, our scheduler learns to determine when to engage specific thinking activities -- such as recalling memories, summarizing experiences, or strategic planning -- based on temporal patterns and historical context. This functional approach allows cognitive modules to be dynamically added or removed without structural reengineering. Meanwhile, we propose a meta-learning strategy for continual policy adaptation, where the scheduler optimizes its cognitive strategy over time using historical interaction logs. Evaluation results demonstrate that our approach effectively learns to schedule cognitive activities based on historical data and can autonomously integrate new thinking modules.

Grammatical error correction (GEC) and explanation (GEE) have made rapid progress, but real teaching scenarios also require \emph{learner-friendly pedagogical feedback} that is actionable, level-appropriate, and encouraging. We introduce \textbf{SPFG} (\textbf{S}poken \textbf{P}edagogical \textbf{F}eedback \textbf{G}eneration), a dataset built based on the Speak \& Improve Challenge 2025 corpus, pairing fluency-oriented transcriptions with GEC targets and \emph{human-verified} teacher-style feedback, including preferred/rejected feedback pairs for preference learning. We study a transcript-based Spoken Grammatical Error Correction (SGEC) setting and evaluate three instruction-tuned LLMs (Qwen2.5, Llama-3.1, and GLM-4), comparing supervised fine-tuning (SFT) with preference-based alignment (using DPO and KTO) for jointly generating corrections and feedback. Results show that SFT provides the most consistent improvements, while DPO/KTO yield smaller or mixed gains, and that correction quality and feedback quality are weakly coupled. Our implementation is available at https://github.com/Skywalker-Harrison/spfg.

Generalized Category Discovery (GCD) leverages labeled data to categorize unlabeled samples from known or unknown classes. Most previous methods jointly optimize supervised and unsupervised objectives and achieve promising results. However, inherent optimization interference still limits their ability to improve further. Through quantitative analysis, we identify a key issue, i.e., gradient entanglement, which 1) distorts supervised gradients and weakens discrimination among known classes, and 2) induces representation-subspace overlap between known and novel classes, reducing the separability of novel categories. To address this issue, we propose the Energy-Aware Gradient Coordinator (EAGC), a plug-and-play gradient-level module that explicitly regulates the optimization process. EAGC comprises two components: Anchor-based Gradient Alignment (AGA) and Energy-aware Elastic Projection (EEP). AGA introduces a reference model to anchor the gradient directions of labeled samples, preserving the discriminative structure of known classes against the interference of unlabeled gradients. EEP softly projects unlabeled gradients onto the complement of the known-class subspace and derives an energy-based coefficient to adaptively scale the projection for each unlabeled sample according to its degree of alignment with the known subspace, thereby reducing subspace overlap without suppressing unlabeled samples that likely belong to known classes. Experiments show that EAGC consistently boosts existing methods and establishes new state-of-the-art results. Code is available at https://haiyangzheng.github.io/EAGC.

We present a unified system addressing both Subtask 3 (answer generation) and Subtask 4 (evidence sentence alignment) of the ArchEHR-QA Shared Task. For Subtask 3, we apply two-stage Quantised Low-Rank Adaptation (QLoRA) to Qwen3-4B loaded in 4-bit NF4 quantisation: first on 30,000 samples from the emrQA-MedSQuAD corpus to establish clinical domain competence, then on the 20 annotated development cases to learn the task-specific output style. Our system achieves an overall score of 32.87 on the official test-2026 split (BLEU = 9.42, ROUGE-L = 27.04, SARI = 55.42, BERTScore = 43.00, AlignScore = 25.28, MEDCON = 37.04). For Subtask 4, we develop a weighted ensemble of three retrieval methods - BM25 with relative thresholding, TF-IDF cosine similarity, and a fine-tuned cross-encoder - to identify note sentences supporting a given gold answer, achieving a micro-F1 of 67.16 on the 100-case test set. Experiments reveal that both subtasks expose the same fundamental challenge: 20 annotated training cases are insufficient to distinguish relevant from irrelevant clinical sentences, pointing to data augmentation as the highest-leverage future direction.

Large Language Models (LLMs) are essential for analyzing and addressing vulnerabilities in cybersecurity. However, among over 200,000 vulnerabilities were discovered in the past decade, more than 30,000 have been changed or updated. This necessitates frequent updates to the training datasets and internal knowledge bases of LLMs to maintain knowledge consistency. In this paper, we focus on the problem of knowledge discrepancy and conflict within CVE (Common Vulnerabilities and Exposures) detection and analysis. This problem hinders LLMs' ability to retrieve the latest knowledge from original training datasets, leading to knowledge conflicts, fabrications of factually incorrect results, and generation hallucinations. To address this problem, we propose an innovative two-stage framework called CRVA-TGRAG (Conflict Resolution in Vulnerability Analysis via Teacher-Guided Retrieval-Augmented Generation). First, to improve document retrieval accuracy during the retrieval stage, we utilize Parent Document Segmentation and an ensemble retrieval scheme based on semantic similarity and inverted indexing. Second, to enhance LLMs' capabilities based on the retrieval of CVE dataset in generation stage, we employ a teacher-guided preference optimization technique to fine-tune LLMs. Our framework not only enhances the quality of content retrieval through RAG but also leverages the advantages of preference fine-tuning in LLMs to answer questions more effectively and precisely. Experiments demonstrate our method achieves higher accuracy in retrieving the latest CVEs compared to external knowledge bases. In conclusion, our framework significantly mitigates potential knowledge conflicts and inconsistencies that may arise from relying solely on LLMs for knowledge retrieval.

Romanized Nepali, the Nepali language written in the Latin alphabet, is the dominant medium for informal digital communication in Nepal, yet it remains critically underresourced in the landscape of Large Language Models (LLMs). This study presents a systematic benchmarking of linguistic adaptation across three comparable-sized open-weight models: Llama-3.1-8B, Mistral-7B-v0.1, and Qwen3-8B. We evaluate these architectures under zero-shot and fine-tuned settings using a curated bilingual dataset of 10,000 transliterated instruction-following samples. Performance is quantified across five metrics spanning seven measurement dimensions: Perplexity (PPL), BERTScore, chrF++, ROUGE-1, ROUGE-2, ROUGE-L, and BLEU, capturing fluency, phonetic consistency, and semantic integrity. Models were fine-tuned using Quantized Low-Rank Adaptation (QLoRA) with Rank-Stabilized LoRA (rsLoRA) at rank r=32 on dual NVIDIA Tesla T4 GPUs, training only approximately 1% of each model's parameters in under 27 total GPU-hours. At zero-shot, all three models fail to generate Romanized Nepali, each exhibiting a distinct architecture-specific failure mode. Following fine-tuning, all three resolve these failures and converge to BERTScore approximately 0.75 and chrF++ greater than 23. Overall dimension-wise assessment across ten criteria identifies Qwen3-8B as the overall recommended architecture, being the only model to produce semantically relevant zero-shot output and leading all structural alignment metrics post-SFT. The adaptation headroom hypothesis is confirmed: Llama-3.1-8B, despite its weakest zero-shot baseline, achieves the largest absolute fine-tuning gains in PPL (Delta = -49.77) and BERTScore (Delta = +0.3287), making it the preferred choice for iterative low-resource development pipelines. This work establishes the first rigorous baseline for Romanized Nepali adaptation in comparable-sized open-weight LLMs.

Retrieval-Augmented Generation (RAG) grounds Large Language Models (LLMs) in external knowledge but often suffers from flat context representations and stateless retrieval, leading to unstable performance. We propose Stateful Evidence-Driven RAG with Iterative Reasoning, a framework that models question answering as a progressive evidence accumulation process. Retrieved documents are converted into structured reasoning units with explicit relevance and confidence signals and maintained in a persistent evidence pool capturing both supportive and non-supportive information. The framework performs evidence-driven deficiency analysis to identify gaps and conflicts and iteratively refines queries to guide subsequent retrieval. This iterative reasoning process enables stable evidence aggregation and improves robustness to noisy retrieval. Experiments on multiple question answering benchmarks demonstrate consistent improvements over standard RAG and multi-step baselines, while effectively accumulating high-quality evidence and maintaining stable performance under substantial retrieval noise.

In large-scale construction projects, the continuous evolution of decisions generates extensive records, most often captured in meeting minutes. Since decisions may override previous ones, professionals often need to reconstruct the history of specific choices. Retrieving such information manually from raw archives is both labor-intensive and error-prone. From a user perspective, we address this challenge by enabling conversational access to the whole set of project meeting minutes. Professionals can pose natural-language questions and receive answers that are both semantically relevant and explicitly time-annotated, allowing them to follow the chronology of decisions. From a technical perspective, our solution employs a Retrieval-Augmented Generation (RAG) framework that integrates semantic search with large language models to ensure accurate and context-aware responses. We demonstrate the approach using an anonymized, industry-sourced dataset of meeting minutes from a completed construction project by a large company in Belgium. The dataset is annotated and enriched with expert-defined queries to support systematic evaluation. Both the dataset and the open-source implementation are made available to the community to foster further research on conversational access to time-annotated project documentation.

We introduce SAGE Celer 2.6, the latest in our line of general-purpose Celer models from SAGEA. Celer 2.6 is available in 5B, 10B, and 27B parameter sizes and benefits from extensive architectural modifications and further pre-training on an undisclosed model. Using our Inverse Reasoning (IR) pipeline, SAGEA natively trains Celer 2.6 to validate its own logic paths, minimizing cascading error and hallucination in complex reasoning tasks. Celer 2.6 also boasts natively integrated multimodal functionality with an end-to-end vision encoder to avoid common pitfalls in adapter-based approaches. Celer 2.6 provides highly competitive results on mathematics, coding, and general intelligence benchmarks (ACUMEN), along with low latency. Most importantly, Celer 2.6 is specifically optimized for South Asian language support, with a custom tokenizer for the Devanagari script and strong performance in both Nepali and Hindi without sacrificing English reasoning ability.