Search Results (776)

Underground coal-mine equipment operates under broadband noise, high dust, humidity, and methane. Acoustic sensing is uniquely suited to this environment: it captures vibration, friction, and airflow signatures without physical contact, incurs low sensor-deployment cost, responds at millisecond speed, and remains effective in low-light, high-dust conditions where optical and vibration alternatives fail. Acoustic fault perception is therefore critically important in underground coal-mine operations. Three unresolved challenges remain: (i) motor whine, material-collision impacts, and ventilation-fan roar compound into a low-SNR soundscape where conventional models lose noise robustness; (ii) acoustic signatures vary widely across equipment types and fault-development stages; and (iii) existing supervised classifiers, trained on imbalanced data, exhibit limited generalization and output only binary judgments, lacking the semantic descriptions that maintenance crews actually need. To address all three, we propose a domain-adaptive audio LLM coupling a BEATs encoder (frozen during Stage II, adapted via Low-Rank Adaptation (LoRA) during Stage I), a Querying Transformer (Q-Former) alignment layer with Dynamic Acoustic Token Compression (DATC), a LLaMA-3.1-8B decoder adapted via LoRA, and Constrained Decoding for Structured Fault Description (CD-SFD) enforcing a three-slot output of fault type, danger level, and handling recommendation. DATC allocates query budget by signal energy to suppress noise-dominated frames; CD-SFD is a state-machine decoder that guarantees the three-slot schema. We release CMEASD: 1200 recordings comprising 24 physical machines (4 per equipment type, 12/6/6 machine-disjoint split). Under machine-disjoint evaluation, the model reaches Macro Accuracy 84.6 ± 1.5% and Macro F1 82.9 ± 1.6%, outperforming the strongest discriminative baseline (PANNs-Transformer, 81.7%) by +2.9 pp and SALMONN-LoRA by +2.5 pp. Ablations attribute +2.4/+1.4/+0.9 pp to DATC, CD-SFD, and the domain prompt. A 30-day mine trial achieves 7.0 s end-to-end latency with 21/30 days of stable, zero-false-shutdown operation. Full article

In the digital education system, the increasing demand for measuring lexical and semantic text similarity has led to the need for developing scalable and intelligent automated grading systems. Traditional assessment methods used for short-answer grading are labor-intensive, time-consuming and prone to human-evaluator bias. Furthermore, computer science curricula have different evaluation challenges as programming-centric courses require rigorous syntactical and structural validation, whereas theoretical courses require deep conceptual and semantic comprehension. To address these dual challenges, we introduce the Virtual University Automatic Short-Answer Grading (VUASAG) framework, a model addressing both statistical-based lexical analysis and Transformer-based semantic architectures. Utilizing an expanded Mohler dataset spanning multiple computer science domains (including Data Structures, Introduction to Programming, Object-Oriented Programming, and Software Engineering), we evaluated the efficiency of traditional string-matching metrics alongside state-of-the-art Transformer-based models (T5, BERT, XLNet, and SBERT). The empirical results demonstrate that token-based statistical models perform robustly on keyword-dependent syntax validation while Transformer-based models excel by capturing deep conceptual context. Specifically, Sentence-BERT (SBERT) achieved superior predictive accuracy across courses, yielding a minimum Root Mean Square Error (RMSE) of 0.9511 for the Introduction to Programming course, paired with moderate positive Pearson and Spearman correlation coefficients. Confidence intervals are estimated using Fisher’s z-transformation with a 95 percentile confidence level. Question-level five-fold and two-fold cross-validation is conducted for all four courses, and SBERT is found to perform better. Finally, to demonstrate the practical utility, we deployed a web-based portal for students’ evaluation leveraging a LLaMA-based pre-trained model to deliver real-time, personalized pedagogical feedback measuring code functionality, quality, documentation, and error handling. This reduces the educator workload while preserving grading integrity. Full article

Background: Large language models (LLMs) can produce clinically plausible recommendations that are not adequately supported by authoritative evidence. Objectives: We introduce GuidelineGuard, a modular multi-agent retrieval-augmented generation pipeline in which a separate Auditor verifies claim–sentence support before a recommendation is surfaced. Methods: The original evaluation used a 73-sentence guideline corpus and GG-Bench-60, with replication across three open-weight backbones. In response to reviewer concerns about benchmark size and selective evaluation, we added a source-traceable GG-Bench-200 stress test and the complete 500-case held-out PQA-L test split of PubMedQA. The revision experiments compare single-pass RAG, a paired multi-agent no-Auditor ablation, and GuidelineGuard; the paired runner is designed to share the Planner–Retriever–Clinician draft so that the Auditor is the only intervention. Checkpoint verification confirmed an identical observable pre-audit state for all 200 GG-Bench cases and 496/500 PubMedQA cases; four PubMedQA cases were regenerated after quota-interrupted resumption and were correct commitments in both arms. Because the originally used hosted Llama endpoints became unavailable after the initial experiments, the expanded runs use openai/gpt-oss-20b for generation and openai/gpt-oss-120b for the Auditor. Results: On GG-Bench-200, single-pass RAG achieved 0.970 operational accuracy, while the no-Auditor and GuidelineGuard arms achieved 0.955 and 0.925, respectively. GuidelineGuard committed on 186/200 cases (coverage 0.930) and was correct on 185/186 commitments (selective accuracy 0.995); all 186 commitments cited at least one gold evidence identifier. Relative to the paired no-Auditor arm, the gate rejected six otherwise-correct commitments and no incorrect commitment. On PubMedQA-500, single-pass RAG achieved 0.644 operational accuracy at 0.950 coverage, the no-Auditor arm 0.638 at 0.896 coverage, and GuidelineGuard 0.550 at 0.736 coverage. Selective accuracy increased across those operating points from 0.678 to 0.712 to 0.747. Within the 496 PubMedQA cases with verified-identical observable pre-audit state, the gate rejected 36 incorrect and 45 correct pre-audit commitments, demonstrating both error enrichment and a substantial false-rejection cost. Conclusions: The expanded results support GuidelineGuard as a selective claim–evidence verification mechanism, not as a universally more accurate generator. Its value is the explicit, auditable coverage–risk trade-off; the appropriate verification threshold is task- and cost-dependent and requires prospective clinical validation. Full article

The convergence of Large Language Models (LLMs) with the Internet of Things (IoT) is driving a transformative shift toward a continuous, context-aware smart healthcare ecosystem. Due to its novelty, existing research in this domain remains fragmented, leaving a critical gap in unified frameworks that synthesize domain applications, functional AI deployment roles, network architectures, and security boundaries. Following PRISMA 2020 guidelines, this paper presents a systematic literature review and quantitative analysis evaluating a selected corpus of 61 peer-reviewed and 14 preprint papers in this domain. Methodologically, we assess a novel hybrid article discovery strategy, finding that an AI-powered prompt-based literature search strategy achieves higher precision than traditional keyword-based Boolean queries (86% vs. 42%) on the evaluated search sample, which may reduce screening workloads. We found that the major limitation of AI-based literature search is non-determinism, which is also an inherent property of LLM-powered applications. To address this, we propose methodological guidelines for using an AI-assisted hybrid literature search strategy. Based on the selected literature, we establish a multi-layer taxonomy organizing the IoT-LLM advances in the healthcare domain across four pillars: application domain, LLM role, IoT device type, and architectural deployment pattern. Quantitative synthesis reveals a heavy research concentration in remote patient monitoring and personal health management (representing 59% of the corpus combined), primarily driven by the data accessibility of wearable sensors (64%). Cross-tabulation uncovers a distinct capability–constraint spectrum: cloud-based deployments lean on heavyweight state-of-the-art models (mainly GPT-family models) for complex semantic reasoning, whereas edge, federated, and blockchain-based hybrid systems leverage localized models (BERT and LLaMA families). Patient data privacy and reduced communication overhead were among the main reasons for choosing localized models. Crucially, our assessment reveals a pervasive neglect of LLM-specific vulnerabilities such as prompt injection and jailbreak attacks and a tendency to treat regulatory frameworks (e.g., HIPAA, GDPR) as design features rather than empirically validated compliance metrics. Finally, we propose an actionable future research agenda prioritizing multi-device system orchestration, emergency care integration, privacy-preserving LLMs, and deployment-scale clinical validation. Full article

In real-world wildfire monitoring, haze, glare, refraction artifacts, and related visual effects can cause false alarms and missed detections. This study explores two implementations: ContextFireVLM and ContextFireAgent. ContextFireVLM fine-tunes Llama-3.2 Vision with LoRA to improve the base model’s ability to identify and analyze wildfires; ContextFireAgent equips an agent with YOLO as a tool to improve the same capabilities. The first and second assessors independently inspect the full image, and the second assessor invokes YOLO only when it reports uncertainty; when their judgments agree, the final reviewer preserves the agreed decision. Because the two implementations use different models, inputs, and training pipelines, these results do not establish a causal comparison or general superiority between LoRA adaptation and the agent design. Full article

Skin biopsy is one of the most valuable diagnostic procedures in dermatology, particularly when clinical findings alone are insufficient to establish a diagnosis. However, obtaining an accurate histopathological diagnosis depends on multiple steps, and errors at any stage of the biopsy pathway may compromise the final result. This review synthesizes current evidence and available guidelines on best practices for skin biopsy, integrating the practical experience of four internationally recognized dermatopathologists to address areas where evidence is limited or poorly standardized. The review covers biopsy planning, selection of the optimal biopsy site and technique, specimen handling and fixation, grossing and laboratory processing, prevention of technical artifacts, the use of ancillary diagnostic techniques, and clinicopathological correlation, with particular attention to challenging anatomical sites and complex diseases. Diagnostic accuracy depends on obtaining a representative specimen, maintaining high technical standards throughout tissue processing, providing adequate clinical information, and ensuring close communication between the clinician and the dermatopathologist. In selected cases, multidisciplinary review is required to reach a definitive diagnosis. Adherence to these principles can optimize diagnostic yield, reduce avoidable errors, and ultimately improve patient care. Full article

Large language models (LLMs) incur prohibitive computational costs when deployed as monolithic systems for multi-domain query processing. This paper proposes Accelerated LLM, a modular architecture that replaces a single general-purpose LLM with an ensemble of task-specialised small language models (SLMs) governed by a neural query router and a Mamdani fuzzy inference system. The router embeds each user query using a frozen sentence encoder and classifies it across four task domains—summarisation, translation, question answering, and text generation—routing confident queries directly to the corresponding SLM. Ambiguous queries are escalated to a three-input fuzzy logic system operating on Query Length, inter-Domain Overlap Score, and Classifier Confidence, enabling principled handling of imprecise inputs. A reinforcement-learning feedback loop, validated through a controlled pilot deployment, continuously refines the routing policy. The complete pipeline, including the sentence encoder, totals approximately 2.14 billion parameters—a 98.8% reduction relative to GPT-3.5 (175 B). The integration of fuzzy logic into the routing stage raises classification accuracy from 91.5% to 94.3% and reduces the hallucination rate to 9.8% (minor) and 6.4% (major). Evaluated on healthcare-augmented benchmarks against ChatGPT-3.5, Claude, Mistral 70B, and two contemporary compact models (GPT-4o-mini and Llama 3.1-8B-Instruct), Accelerated LLM achieves competitive or superior task-specific performance at a fraction of the parameter count. A small-scale pilot evaluation in the legal domain indicates that the routing and fuzzy logic components retain partial effectiveness beyond the primary healthcare setting, though full multi-domain validation remains future work. Full article

EASE-CloudNet is a two-phase safety-alignment framework for generative small language models (SLMs) deployed on resource-constrained edge nodes. Its input is a natural-language user query and its output is a safe, helpful natural-language response or refusal; network-traffic classification and resource-scheduling actions are outside the task evaluated in this study. In Phase 1, a cloud teacher uses a security policy graph to generate structured safety rationales and response targets, which are distilled into Qwen2.5-1.5B/3B and Llama3.2-3B students. In Phase 2, an offline heterogeneous graph and a two-layer GraphSAGE model identify vulnerable semantic regions; these vulnerability targets supervise a lightweight edge-side router. We formulate deployment cost as a differentiable gate-conditioned expectation, so measured latency and energy constants affect the router through its reasoning probability. In the Qwen2.5-1.5B ablation experiments, the full model obtains 3.9% StrongREJECT ASR, 54.7% MMLU accuracy, and 70 average generated tokens; an A100 reference profile reports 18.8 ms/query and 2.37 J/query, or 3.3% latency and 2.6% measured GPU-energy overhead over the unaligned model. Physical edge runs measured a direct/reasoning end-to-end latency of 41.2/68.7 ms on Jetson Orin NX and 62.5/105.3 ms on Snapdragon 8 Gen 3, with a direct/reasoning energy of 0.48/0.79 and 0.71/1.18 J/query, respectively. Equal-seed Holm–Bonferroni-corrected tests confirm lower ASR than EASE on StrongREJECT and WildJailbreak for all three base models ($p<0.01$). Full article

Retrieval-augmented generation (RAG) systems for scientific literature require evidence-based choices of document segmentation, representation, retrieval, and generation components, particularly when the source collection varies in topical specificity and document structure. This study addresses the lack of an end-to-end, component-level comparison of these choices for soil science question answering. A three-stage evaluation was conducted across general, domain-specific, and geospatial soil science corpora. The corpus combines foundational soil science books, peer-reviewed research articles, European soil monitoring material, and geospatial mapping publications, thereby covering both broad disciplinary concepts and specialized scientific evidence. The study compares four chunking strategies, three embedding models, five retrieval methods, and five large language models. In Experiment 1, semantic chunking with text-embedding-3-large achieved the highest aggregate retrieval scores (recall@1 = 0.824; MRR = 0.819), whereas text-embedding-3-small delivered practically comparable performance at lower cost. In Experiment 2, hybrid reciprocal rank fusion achieved recall@5 values of 0.957, 0.960, and 0.647 for the general, domain-specific, and geospatial corpora, respectively; the cross-encoder reranker showed weaker rank quality on scientific content. In Experiment 3, model responses attained BERTScore values of 0.909–0.927 and faithfulness of at least 0.993; these automated measures indicate low contradiction with retrieved context but do not establish answer completeness or human-perceived correctness. The study provides a reproducible component-level evaluation design, characterizes the effect of corpus specificity on RAG retrieval, and identifies a practical configuration for soil science literature retrieval. Among the models retained for direct aggregate comparison, Llama 3.1 8B offered the most favorable observed balance of answer quality, latency, cost, and model openness. Full article

This paper presents a comparative benchmarking study evaluating the regulatory comprehension of four open-source large language models, TinyLlama-1.1B-Chat, Gemma-2B-Instruct, Llama-3.1-8B-Instruct, and Mistral-7B-Instruct-v0.3, on the EU Artificial Intelligence Act (Regulation EU 2024/1689). A custom benchmark of 100 multiple-choice questions was constructed and classified across three dimensions, namely cognitive difficulty, knowledge type, and regulatory mechanism, with a balanced answer key distribution. Each model was evaluated under vanilla inference, relying exclusively on parametric knowledge, and retrieval-augmented generation (RAG), in which relevant Act passages are retrieved using a FAISS-indexed sentence embedding pipeline with standardized top three chunk retrieval applied uniformly across all models. Vanilla accuracy ranges from 22.0% for TinyLlama-1.1B to 80.0% for Llama-3.1-8B, indicating that larger models perform better. Under the retrieval configuration employed, RAG improves performance of the evaluated models, with gains from 3.0 percentage points for TinyLlama-1.1B to 17.0 for Gemma-2B and 10.0 for both 7B–8B models. Questions related to procedural knowledge, governance and enforcement emerge as weaknesses in the baseline, vanilla setting, partially mitigated by retrieval. Qualitative analysis identifies two RAG failure modes: retrieval failures, where the embedding mechanism returns informationally insufficient passages, and integration failures, where the correct passage is retrieved but not correctly exploited. The study contributes an empirical characterization of small and medium language model regulatory comprehension and a reusable 100-question benchmark. Full article

The advent of large language models and Retrieval-Augmented Generation (RAG) models has greatly enhanced intelligent information systems. This has resulted in the development of context-aware and knowledge-grounded response generation. This has been highly beneficial in the context of religious advisory systems, which require precision, correctness, and knowledge grounding. For Islamic rituals like Hajj and Umrah, the user needs precise and accurate procedures to follow, which must adhere to specific sequences and knowledge grounding. However, the existing models have many limitations in this context, like hallucinations, a lack of procedural knowledge, bilingual inconsistencies, and an inability to incorporate safety constraints. This has made these models unsuitable for contexts in which incorrect responses can have serious implications. Therefore, in this context, this paper proposes a Deterministic Procedure-Aware bilingual Retrieval-Augmented Generation (DPAM-RAG) model, which can be highly beneficial in designing religious advisory systems. The proposed model can be highly beneficial in designing religious advisory systems. The proposed model integrates dataset modeling, procedure-aware chunking, bilingual alignment, and deterministic transformer-based response generation. Additionally, a confidence-based refusal strategy has been proposed to avoid the generation of responses that can be considered incorrect or out of context. The proposed model has been tested through an extensive experimental setup, which includes multiple transformer models like GPT, LLaMA-2, Mistral, MPT, and BLOOMZ. The experimental results have shown promising outcomes, which can be considered highly beneficial in designing trustworthy AI models. Full article

Fault diagnosis in distributed systems is challenged by complex service dependencies, cascading anomaly propagation, and similar symptom patterns. This paper proposes a knowledge-enhanced iterative reasoning framework that integrates large language models (LLMs) with a numerical domain knowledge graph (KG). The KG encodes fault–symptom relations, anomaly directions, and training-derived mean and standard-deviation intervals. Structured prompting first generates candidate faults; interval verification then rejects numerically inconsistent candidates. For retained candidates, counterfactual reasoning constructs hierarchical causal chains, KG traversal refines missing or inconsistent links, and a deterministic evidence score supports acceptance, exclusion, early stopping, and fallback across at most five iterations. Under the common 68-case evaluation protocol for eight known single-root-cause faults in the controlled Redis-based testbed, the complete framework achieved 100.00% Accuracy, Macro-F1, and Balanced Accuracy with GPT-4o and GPT-5.2, compared with 91.18% accuracy for KG-only reasoning and 85.29–89.71% for Random Forest, XGBoost, and Transformer baselines. GPT-3.5 reached 98.53%, whereas LLaMA-3.1-8B reached 80.88%, showing that the incremental KG–LLM gain is backbone-dependent. Five GPT-4o repetitions and three GPT-5.2 repetitions yielded 100.00% ± 0.00, and all three metrics remained at 100.00% across the evaluated Z-score thresholds, iteration limits, and interval tolerances. The framework therefore provides highly accurate, stable, and traceable diagnoses within the evaluated Redis-based distributed-service protocol, while providing explicit intermediate reasoning and solution retrieval. Full article

Background/Objectives: Emergency department (ED) overcrowding contributes to delayed care, prolonged length of stay (LOS), resource strain, and adverse patient outcomes. This systematic review aimed to examine how artificial intelligence (AI) and machine learning (ML) have been used to address ED crowding and patient flow, with emphasis on modeling approaches, validation practices, and real-world implementation. Methods: Following PRISMA 2020 guidelines, Scopus, Embase, Ovid MEDLINE, and CENTRAL were searched for relevant studies published from 2020 onward. After deduplication, 1888 records underwent title and abstract screening using two locally deployed LLaMA models with human adjudication. Screening performance was assessed against 150 manually annotated records. Full-text eligibility assessment and structured data extraction were conducted independently by multiple reviewers, with disagreements resolved by consensus. Results: Thirty-two studies were included. Most were retrospective, single-site investigations using electronic health record, administrative, or operational data. Common outcomes included ED LOS, waiting time, occupancy, boarding, disposition, and crowding indices. Tree-based and boosting models frequently performed well, although no approach was consistently superior across tasks and settings. Most studies relied on same-site validation, while external and temporal validation were uncommon. Prospective implementation, workflow integration, model maintenance, and direct operational, clinical, economic, or equity impacts were rarely evaluated. For LLM-assisted screening, LLaMA 4 Scout achieved 84.0% accuracy, 80.0% recall, 88.9% precision, and an F1 score of 84.2%, compared with 78.0%, 67.5%, 88.5%, and 76.6%, respectively, for LLaMA 3.3 on 150 randomly sampled papers. Conclusions: AI and ML show promise for addressing ED overcrowding, but the literature remains concentrated at the model-development stage. Future research should prioritize standardized outcomes, multicenter validation, prospective implementation, and direct evaluation of operational and patient-care outcomes. Full article

Alpacas have gained increasing popularity in Poland; however, no official national standards for their husbandry have yet been established. This study aimed to evaluate alpaca husbandry practices on Polish farms using an anonymous questionnaire. A total of 44 farms were included in the analysis, all of which kept alpacas, while only 9.1% also maintained llamas. A high level of adherence to preventive healthcare practices was observed, with regular deworming implemented on 90.9% of farms and vaccination reported by all respondents. Dietary supplementation was also widely practised, with vitamin D being the most frequently administered supplement (90.9%). However, the assessment of biosecurity measures revealed important deficiencies in epidemiological risk management, as only 59.3% of owners quarantined newly introduced animals before integrating them into the herd. These findings provide the first comprehensive overview of alpaca husbandry practices in Poland and may serve as a basis for future research and the development of official husbandry standards for alpacas. Full article

Serving large language models (LLMs) on cloud spot and preemptible instances reduces costs by 60 to 90 percent compared to on-demand pricing, but unpredictable instance preemptions cause request failures, KV-cache state loss, and degraded user experience. We present MorphCloud-LLM, an elastic LLM serving system designed to achieve the reliability properties of on-demand serving at spot-instance pricing. MorphCloud-LLM integrates three synergistic components: (1) an asynchronous incremental KV-cache checkpointing engine that streams only delta state to disaggregated persistent storage with less than 3% throughput overhead, enabling sub-second KV-cache delta streaming and reconstruction for KV-cache sizes up to 32 GB on replacement instances (total end-to-end migration latency: 1390 ms); (2) a gradient-boosted preemption prediction model trained on spot market telemetry that achieves 89% recall at a 30-s prediction horizon, providing sufficient lead time for proactive migration before forced eviction; and (3) a speculative decoding continuity engine that offloads draft model token generation to on-demand fallback nodes during migration windows, bounding the user-visible interruption to a sub-second buffering pause. MorphCloud-LLM is deployed and evaluated on AWS and GCP using LLaMA-70B and Mixtral-8x7B across 521 trace-injected preemption events, achieving up to 76% cost reduction under active-serving accounting (69.8% for LLaMA-70B; 67% including warm standby fallback capacity) with only 2.1% p99 latency overhead and zero dropped requests. Extensive ablation studies confirm the contribution of each component to overall system resilience. Note that preemption events are reproduced via a trace-driven simulation framework built on empirical AWS and GCP spot interruption traces rather than fully uncontrolled live production preemptions. Production generalizability under uncontrolled preemption—including simultaneous multi-node failures, network congestion, storage contention, and replacement-instance scarcity remains subject to future validation in sustained live deployments. Full article