The optimization of complex manufacturing processes, such as Chemical Vapor Deposition, requires integrated approaches that combine physical modeling with advanced data-driven methodologies. This review synthesizes recent advances in hybrid modeling frameworks that merge equation-based computational fluid dynamics, machine learning, and natural language processing models to enhance process understanding, prediction, optimization and control. In particular, natural language processing techniques are leveraged to generate embedding-based predictors that inform learning tasks. The proposed framework integrates data acquisition, dimensionality reduction, and feature engineering with contextual language processing embeddings, surrogate modeling, and sensitivity analysis. This results in improved forecasting accuracy and interpretability. Key applications include coating thickness prediction, process regime classification, and critical parameter identification using SHAP analysis and Sobol’ indices. Nevertheless, significant challenges remain, including limitations in sensor infrastructure, assessment of dataset sufficiency for specific industrial objectives, and restricted generalizability across reactor designs. This work highlights how hybrid frameworks, together with natural language processing models applied to industrial process datasets, can bridge the gap between data availability in industrial environments and the actionable insights required for practical implementation, while identifying necessary future directions for robust, scalable, and interpretable modeling systems in advanced manufacturing. • Hybrid framework that unifies CFD, machine learning, and NLP for industrial CVD. • Dimensionality reduction and surrogate models accelerate predictive CVD analytics. • Clustering and feature-importance analysis reveal dominant process mechanisms. • NLP-based encoding enhances predictive modeling of categorical industrial variables.

Comparative effectiveness of recreational therapies on cognition, neuropsychiatric symptoms, and psychosocial outcomes in individuals with mild cognitive impairment and dementia: A network meta-analysis.

Dyslexia is a common learning disability affecting language processing and literacy acquisition. In China, although prevalence rates are comparable to Western countries, awareness among teachers remains limited. Teachers are key to the early identification and support of students with dyslexia, yet many feel unprepared or hold misconceptions about the condition. This quantitative survey included 909 mainstream primary and secondary school teachers from Guangdong Province, China. A structured questionnaire assessed teachers' knowledge of developmental dyslexia, attitudes toward affected students and self-efficacy in supporting them. Teachers generally lacked sufficient knowledge of dyslexia and held notable misconceptions about its causes and interventions. Many expressed uncertainty about their ability to support students with dyslexia and showed ambivalent attitudes regarding these students' potential for improvement. Correlation and regression analyses indicated that teachers' knowledge and attitudes were significantly associated with and positively predicted their self-efficacy in teaching students with dyslexia. These findings reveal significant gaps in teachers' knowledge and confidence related to dyslexia, which may hinder effective inclusive education. The results highlight the need for targeted teacher training and professional development to address misconceptions, foster positive attitudes and enhance self-efficacy, ultimately improving support for students with dyslexia in mainstream classrooms.

We study and measure the uncertainty in the Spanish version of the minutes of the meetings of the Governing Board of the Bank of Mexico and relate it to key monetary policy variables. These minutes summarize the information about the domestic and international economic, inflation and financial background, as well as the reasoning and rationale behind the chosen monetary policy decision. In particular, we conceive various uncertainty indices using unsupervised machine learning techniques for natural language processing and a large language model. A first set of uncertainty indices is constructed by exploiting latent Dirichlet allocation (LDA), whereas a second set is based on word embedding (implemented with the skip-gram (SG) model) and k-means. Finally, a third set of uncertainty indices is based on ChatGPT (GPT). For each of these three approaches, we create an uncertainty index for the whole set of minutes, and three section-specific uncertainty indices for the three main sections of the minutes. Then, we obtain an overall Monetary Policy Uncertainty ( MPU ) index and three section-specific indices by averaging the corresponding LDA, SG, and GPT indices. Thus, with the implementation of an SVAR model, we find that a positive shock in the MPU index is related to an increase in money supply, in the consumer price index, in the target for the overnight interbank interest rate, and to a depreciation of the Mexican peso.

Artificial intelligence in sleep medicine I: Diagnosis, treatment, care, and research.

The art of semantics in radiology reporting.

Searching disease-related genes with Mapping of Biological Entities from Literature (MaBEL).

Tag recommendation systems face persistent semantic challenges, including synonymy, polysemy, and contextual ambiguity, which continue to affect recommendation quality despite progress in addressing sparsity and cold start issues. This paper provides a systematic literature review of tag recommendation methods from 2010 to 2025 and introduces a method-challenge mapping framework that links major recommendation paradigms to the specific semantic and structural problems they are best suited to address. Drawing on insights from this review, we include an illustrative hybrid pipeline that demonstrates how CF, CB, and CA techniques can be combined with lightweight natural language processing components such as GloVe-based semantic clustering, sentiment signals, and contextual word sense handling. This example is intended solely as a demonstration of how survey findings can guide practical system design. Using the MovieLens 20 M dataset, we present a small empirical demonstration that shows how integrating semantic and contextual cues can support improvements in precision, recall, and diversity under realistic constraints. The primary contribution of this work is the systematic survey and analytical framework, with the illustrative example serving as a practical companion for researchers and practitioners.

EfficientLoRA: Rethinking the efficiency of low-rank adaptation in pre-trained language models.

SEEK: A simple defense to model hijacking attack.

Adversarial attacks present significant risks to machine learning (ML) systems, exploiting model vulnerabilities and threatening the integrity, security, and trustworthiness of applications across multiple sectors. This paper provides a comprehensive review of adversarial attack types—white box, black box, and other type of attacks—and examines tailored attacks and defense mechanisms across domains such as Internet of Things (IoT), healthcare, industrial control systems, autonomous vehicles, speech recognition, natural language processing (NLP), finance, and Large Language Models (LLMs). Each domain introduces unique adversarial challenges and demands specific countermeasures, from anomaly detection to adversarial training and robust model architectures. By systematically categorizing both attack methodologies and defense strategies, this survey offers a holistic understanding of adversarial dynamics across fields, highlighting critical areas for further research and the development of resilient, cross-domain ML defenses. • Comprehensive Analysis: Reviews adversarial attacks across multiple data types and their impact on machine learning models. • Taxonomy Development: Proposes a structured taxonomy of adversarial attacks aligned with the MITRE ATLAS framework. • Vulnerability Identification: Identifies vulnerabilities across data modalities exploited by adversarial attacks. • Attack Categorization: Categorizes adversarial attack techniques across different data types and ML systems. • Domain-Specific Taxonomy: Examines adversarial attacks across domains including IoT, healthcare, NLP, speech, and LLMs.

Organizations generate vast amounts of unstructured textual data – a valuable source of information that frequently remains underutilized for process mining. However, textual descriptions often record exceptions and manual activities absent from structured data, and therefore, enable a better understanding of deviations from the expected business process behavior. Importantly, unstructured sources typically retain the object-centric characteristics of real-world processes – information that gets flattened or lost in case-centric event logs. Yet, existing approaches primarily target structured data sources or produce case-centric event logs. To address this gap, we present an automated approach to derive object-centric event logs directly from unstructured textual descriptions. The approach comprises two subcomponents: a collector that identifies events and objects (including their attributes and relationships), and a refiner that consolidates and cleans the extracted information. We instantiate each subcomponent in heuristic and generative implementations and create four pairwise combinations of collector and refiner instances to assess the effectiveness of heuristic natural language processing and generative artificial intelligence techniques. We compare these variants quantitatively and qualitatively in a controlled, artificial setting based on synthesized texts and demonstrate the practical utility on two naturally occurring corpora (fire status updates and a legal judgment). Our results show that the configurations with a generative collector achieve the highest extraction quality. In particular, the fully generative variant produces coherent and standardized event and object labels. Overall, this study fills a notable research gap by enabling the incorporation of textual information into process mining applications. • Proposes an approach to extract object-centric event logs from textual descriptions. • Develops the approach using the Design Science Research methodology. • Implements the approach using heuristic NLP and generative AI techniques. • Evaluates the approach on synthetic and naturalistic textual descriptions. • Confirms the approach’s practical utility on fire status updates and a legal judgment.

Transformer architectures based on the attention mechanism have revolutionized natural language processing (NLP), driving major breakthroughs across virtually every NLP task. However, their substantial memory and computational requirements still hinder deployment on ultra-constrained devices such as wearables and Internet-of-Things (IoT) units, where available memory is limited to just a few megabytes. To address this challenge, we introduce EmbBERT, a tiny language model (TLM) architecturally designed for extreme efficiency. The model integrates a compact embedding layer, streamlined feed-forward blocks, and an efficient attention mechanism that together enable optimal performance under strict memory budgets. Through this redesign for the extreme edge, we demonstrate that highly simplified transformer architectures remain remarkably effective under tight resource constraints. EmbBERT requires only 2 MB of total memory, and achieves accuracy performance comparable to the ones of state-of-the-art (SotA) models that require a 10 × memory budget. Extensive experiments on the curated TinyNLP benchmark and the GLUE suite confirm that EmbBERT achieves competitive accuracy, comparable to that of larger SotA models, and consistently outperforms downsized versions of BERT and MAMBA of similar size. Furthermore, we demonstrate the model's resilience to 8-bit quantization, which further reduces memory usage to just 781 kB, and the scalability of the EmbBERT architecture across the sub-megabyte to tens-of-megabytes range. Finally, we perform an ablation study demonstrating the positive contributions of all components and the pre-training procedure. All code, scripts, and checkpoints are publicly released to ensure reproducibility: https://github.com/RiccardoBravin/tiny-LLM.

While previous studies have focused on how contextual elements influence reading, our research adopts a novel approach to the dynamic interaction between words and their context. Using eye movements during naturalistic Serbian sentence reading, we analyzed how both morphological and semantic information carried by consecutive nouns in a sentence impact reading behavior. We considered a broad range of potential predictors, categorized as either benchmark (word length and frequency), paradigmatic (relative entropy, letter-triplet activation, and competition), or syntagmatic (cosine similarity, contextual similarity, and typicality) to predict four eye movement measures: landing position, first fixation duration, total fixation duration, and average pupil size. A multivariate statistical model that enabled simultaneous testing of multiple predictors against several dependent variables revealed a complex interaction among word properties within context. While previously identified predictors such as length and frequency come out strong, our approach reveals the relational nature of word- and context-specific effects and brings out how both paradigmatic and syntagmatic information shapes language processing. (PsycInfo Database Record (c) 2026 APA, all rights reserved).

From decision support to clinical integration: A scoping review of artificial intelligence in prehospital airway management.

AI literacy among healthcare professionals and students in the Americas.

Artificial intelligence language models for medical text analysis: A systematic review.

ICU patients frequently receive treatments clinicians perceive are futile which can cause conflict between clinicians, patients and families. Medical futility lacks a consensus definition, yet this ambiguous and contentious term is used in medical notes. What themes are associated with futility mentions in ICU notes? How have themes' frequencies changed over time? Mixed methods study of ICU notes (e.g., H&P, progress notes) written by clinicians (e.g., physicians, nurses) for adult patients at a large hospital system from 2010 to 2020. Neural network models identified terms most associated with "futile" or "futility." Distributional semantic analysis grouped terms into themes. Regression modeling explored longitudinal changes in themes' frequencies. Across 2,460,169 notes for 9912 patients, the annual average count of unique notes with futility mentions was 137 per 100,000 and unchanged from 2010 to 2020. 8 themes were identified among terms most associated with the words "futile" or "futility." The most represented themes were Decision Making (annual average 18% [95% CI: 16%-19%]), Assessing, Prognosticating, and End-of-Life Outcomes (15%, 13%-16%), and Identifying Sentiments (13%, 10%-15%). Recording Code Status was the least represented theme in 2010 (4%) and increased over time (9% in 2020, P=0.001). Use of futility was rare and stable across a decade of ICU notes. Semantic analysis indicates clinicians use futility in heterogeneous contexts. Changes in themes' frequencies may reflect clinicians' evolving conceptions of medical futility. These findings could guide development of EHR-based interventions to address perceived futile treatments which contribute to clinicians' moral distress.

The Business Process Modeling Notation (BPMN) is the de facto standard for business process modeling. While widely adopted for its intuitive graphical notation, its execution semantics described in natural language lacks a commonly agreed formal foundation, leading to variability in execution across different BPM systems (BPMSs) and increasing the risk of creating models with semantic errors costly to correct at runtime. Although many formalisms have been used to model portions of BPMN, their reasoning capabilities are mostly restricted to control-flow, making them unsuitable for semantic analysis where data and global exception handling play a central role in execution. To address this, we propose a formalization from BPMN to ConGolog, a logical concurrent processes language based on the Situation Calculus, for representing and reasoning about dynamic domains. A major innovation is using ConGolog to rigorously capture the semantics of BPMN global exceptions. Our framework supports advanced reasoning, allowing for semantic analysis of BPMN models before execution to predict runtime errors within a safe simulation setting, while laying the foundation for reasoning layers in next-generation AI-augmented BPMSs. We validate the approach through a prototype and comprehensive evaluation, demonstrating the computational feasibility of the translation and the semantic correctness of reasoning tasks.

Distinct cross-modal coupling between the written and auditory language networks during reading and spoken word processing.