Endometriosis is a chronic gynecological disease characterized by the growth of endometrial-like tissue outside the uterus, leading to pelvic pain, infertility, and other major health complications. Though some studies have tried to determine the factors that influence endometriosis and its clinical manifestations, the predictive modeling approaches have many limitations. Most of the models are based on complex algorithms that are not interpretable in a clinical setting, and therefore, practitioners cannot easily apply the findings. Moreover, much of the previous research often focuses on specific patient subsets or limited disease stages, leaving critical gaps in understanding the condition comprehensively. The purpose of this study was to address these challenges by using a rule-based model for analyzing various clinical data at each stage of endometriosis. The aim will be to develop interpretable and actionable diagnostic rules that will be able to empower health professionals in better understanding, diagnosis, and management of endometriosis while filling the existing gaps in knowledge within the field. Clinical data were collected and preprocessed to deal with missing values, selection of relevant features, and standardization of variables. The CN2 rule induction algorithm was applied to 1,489 records with 52 clinical variables to generate interpretable classification rules linking clinical variables with the probability of endometriosis. The performance of the model was evaluated using several metrics: accuracy, F1-score precision, recall, and AUC. Derived rules were validated by clinical experts to ensure both statistical rigor and practical relevance. The CN2 model achieved an AUC of 0.906 and a classification accuracy of 0.803 for binary classification. For multi-class classification of disease stages, the model demonstrated an average AUC of 0.705. Key findings were pelvic pain, dysmenorrhea, dyspareunia, severe bleeding, tumor markers, age, and BMI. The rules generated by the CN2 model provided interpretable insights, facilitating better understanding and management of endometriosis. The CN2 rule induction model has been effective in bridging the gaps in endometriosis research by providing interpretable and actionable diagnostic rules. Its integration into clinical practice may improve the precision of diagnosis, reduce diagnostic delays, and offer strategies for treatment, hence possibly improving patient outcomes. Not applicable.

Predictive multiscale cellular modeling is emerging as a consequential direction in precision medicine, converging hypothesis grammars, digital twins, and integrative genomics to interrogate tumor-immune dynamics, therapeutic resistance, and cellular plasticity. This perspective synthesizes recent progress across these domains and critically maps their translational potential alongside their current limitations. Hypothesis grammars translate mechanistic theories into executable agent-based models (ABMs) and hybrid ODE-PDE systems, enabling rapid in silico hypothesis testing while lowering the authoring barrier for domain scientists. Patient-specific digital twins, driven by multi-omics data, employ stochastic ensemble methods to simulate clonal evolution and microenvironmental interactions, though prospective clinical validation of these capabilities remains at an early stage. Integrative genomics, leveraging algorithms such as SCODE and SimiC, infers causal gene regulatory networks (GRNs) using Bayesian variational autoencoders, embedding dynamic intracellular logic into tissue-scale simulations. Emerging applications include in silico oncology trials for optimizing checkpoint blockade and combination therapies. Large language models are being explored to enhance rule induction, while FAIR-compliant digital cell repositories aim to ensure reproducibility and reuse. Verification, validation, and uncertainty quantification (VVUQ) via Sobol sensitivity analysis and Kennedy-O'Hagan calibration are identified as essential components for addressing non-identifiability and supporting regulatory credibility. Federated learning is discussed as a means of mitigating privacy and bias concerns in multi-institutional settings. Together, these converging approaches outline a plausible pathway toward virtual clinical trials and adaptive theranostics, contingent on the prospective validation, data infrastructure, and governance frameworks that clinical deployment will require.

Shoreline change detection from remote sensing imagery remains challenging in environments subject to water level fluctuations, as remotely sensed shoreline positions reflect instantaneous hydrodynamic states rather than true geomorphic change. In the Great Lakes, seasonal and short-term water level variations can produce apparent shoreline shifts unrelated to sediment dynamics. Reliable calibration with bathymetry and water level data can mitigate this effect, but such data are often unavailable or difficult to obtain for many coastal and lacustrine systems worldwide. To address this limitation, we proposed a morphology-based framework that quantifies geometric change between successive shoreline curves using a discrete Fréchet distance, a modified Euclidean distance and a Union distance metric. Rather than relying solely on cross-shore displacements, the approach leverages shape similarity to differentiate water-level-driven shifts from true morphological change. We evaluated the framework across three spatial scales (100 m, 500 m, and 1000 m) along 125 km of southwestern Lake Michigan coastline using 2010 and 2020 aerial imagery, benchmarking against water-level-calibrated DSAS erosion hotspots. The Fréchet distance improved monotonically with scale, achieving strong agreement at 1000 m (F1 = 0.84, Spearman ρ = 0.79) but limited reliability at 100 m. While individual morphology-based metrics appeared competitive with or inferior to uncalibrated DSAS at each scale, the union of both distances substantially outperformed uncalibrated DSAS at management-relevant scales (F1 of 0.64 vs. 0.50 at 500 m and 0.79 vs. 0.42 at 1000 m), reflecting the complementary nature of shape-based and displacement-based detection. The Patient Rule Induction Method (PRIM) further identified gentle nearshore slopes and moderate separation from engineered structures as the geomorphic conditions under which the morphology-based and calibrated erosion indicators converged most closely (in-box F1 = 0.92 at 1000 m and 0.72 at 500 m). These results suggest that the proposed framework, particularly the complementary union of both metrics, provides a practical, calibration-free alternative for multiscale shoreline change screening in lacustrine and microtidal, data-limited environments, while local-scale applications still benefit from explicit water-level correction.

This analysis distinguishes between the narrow optimisation of current AI and the extrapolation needed for Artificial General Intelligence (AGI). Today, AI mainly consists of Artificial Narrow Intelligence (ANI), which is good at pattern recognition within its training data. Architectures like the Transformer, with mechanisms such as self-attention and gradient descent, are designed to identify statistical correlations within a specific distribution, such as spam SMS features. But this makes them fragile; they struggle to apply learned rules to slightly different domains, such as email filtering, because their knowledge is limited to the "convex hull" of the training examples. This isn’t a scale issue but a structural one-so-called "zero-shot" learning is broad interpolation, not true skill. Conversely, biological cognition emphasises extrapolation-forming abstract, hierarchical schemas from limited experience and applying them to new situations. Human learning, like a child moving from crawling to walking, uses embodied cognition to relate concepts like gravity and momentum to physical reality, enabling effective knowledge transfer. Large Language Models are less data-efficient than humans, who learn language through social and physical grounding. Their failure on benchmarks such as the Abstraction and Reasoning Corpus (ARC-AGI), which test abstract rule induction from a few examples, highlights this gap. Public discussions on Brain-Computer Interfaces are often misleading; these systems are advanced, narrow AI applications for signal processing and decoding motor intentions, not for understanding or reading abstract thoughts. Achieving AGI will require a shift beyond merely scaling current architectures, which face data and energy constraints. The future involves integrating causal reasoning, neuro-symbolic systems, and embodied AI to close the gap between statistical pattern matching and genuine, general understanding.

ABSTRACT This paper develops and evaluates a human-centered decision support framework for explainable social media message prioritization. Adopting a Design Science Research (DSR) perspective, it addresses message handling as an ordered multicriteria sorting problem. The proposed approach is operationalized through a two-phase method comprising criteria elicitation, reference-message evaluation, rule induction, iterative validation with the decision-maker and application of the validated rule base to new messages. The framework is applied within a French business school for institutional communication on X. Evaluation combines decision-maker validation, dominance-based consistency analysis and stratified 5-fold cross-validation. The results show that the induced rule base provides an interpretable and operational representation of the decision-maker’s prioritization logic, with a global quality of approximation of 0.86, a mean accuracy of 0.74 and a balanced accuracy of 0.76.

Explaining Performance Issues of Cloud Applications from Logs using Rule Induction and Dempster-Shafer Theory

The work presents a hybrid neuro-symbolic method that combines a large language model (LLM) and a finite-state transducer (FST) to ensure morphological correctness in text generation for agglutinative languages. The system automatically extracts rules from corpus data: for local examples of word forms, the LLM produces sequences of morphological analyses, which are then aggregated and organized into compact descriptions of morphotactic rules (LEXC) and allomorph selection (regex). During generation, the LLM and FST operate jointly: if a token is not recognized by the automaton, the LLM derives a “lemma+tags” pair from the context, and the FST produces the correct surface form. A literary corpus (~1600 sentences) was used as the dataset. For a list of 50 nouns, 250 word forms were extracted. Using the proposed algorithm, the LLM generated 110 context-sensitive regex rules along with LEXC morphotactics, from which an FST was compiled that recognized 170/250 forms (~70%). In an applied machine translation test on a subcorpus of 300 sentences, integrating this FST into the LLM cycle improved quality from BLEU 16.14 / ChrF 45.13 to BLEU 25.71 / ChrF 50.87 without retraining the translator. The approach scales to other parts of speech (verbs, adjectives, etc.) as well as to other agglutinative and low-resource languages, where it can accelerate the development of lexical and grammatical resources.

Neuro-Symbolic Approach to Augmented Text Generation via Automated Induction of Morphotactic Rules

OP 37: Health Status 3, B210 (FCSH), September 5, 2025, 11:30 - 12:30Aims: Migrant populations in Europe often face challenges in achieving optimal blood pressure control. While traditional statistical methods have provided valuable insights, machine learning offers the potential to uncover complex interactions among factors influencing blood pressure control. We therefore applied machine learning techniques to the multi-ethnic HELIUS cohort in the Netherlands to identify subgroups of hypertensive participants achieving optimal blood pressure control based on migration background, phenotypic traits, and individual antihypertensive medications.Methods: Data from 1,580 hypertensive participants were analyzed over seven years (2015–2021). Participants were classified as Dutch (non-migrants) or migrants, including those of South Asian Surinamese, African Surinamese, Turkish, Moroccan, and Ghanaian origins. The dataset included 55 individual antihypertensive medications, along with demographics, lifestyle factors, comorbidities, and treatment adherence. Blood pressure control was defined according to WHO criteria. Subgroup discovery techniques—Patient Rule Induction Method (PRIM), Subgroup Set Discovery ++ (SSD ++), and APRIORI-SD—were used to identify subgroups achieving optimal blood pressure control.Results: Twenty-one subgroups were identified (11 with PRIM, 5 with APRIORI-SD, and 5 with SSD ++). Most subgroups were defined solely by medication variables, with optimal control rates around 70%. However, a notable subgroup was identified using SSD ++: 96% of participants with higher baseline blood pressure who used lercanidipine achieved optimal control, irrespective of migration background, lifestyle, demographics, or other phenotypic traits.Conclusion: Our findings demonstrate that medication type plays a more significant role in achieving blood pressure control than migration background, lifestyle, or comorbidities. Identifying ways to predict individual antihypertensive medication responses before initiating treatment could significantly improve hypertension management, regardless of migration background.

The safe and reliable operation of energy storage stations critically depends on their thermal management systems, specifically the health states or working conditions of involved sensors, such as temperature, humidity, and pressure sensor. Impacted by several environmental factors, some indiscernible defects including signal drift, elevated noise, and response lag may affect the exact surveillance of batteries, leading to potential combustion or even explosion, which requires fault risk early-warning to support timely maintenance. These multi-sensor environmental factor data typically exhibit mixed characteristics, component coupling, and high uncertainty, thus impacting diagnostic accuracy and robustness. With this motivation, this study proposes a pattern-guided framework for vulnerability diagnosis using Component Importance Measure. A pattern-guided strategy is first designed to perform rule induction and fuzzy processing on discrete and continuous sensor data, respectively, to extract underlying vulnerability-related components. Subsequently, a component Importance Measure, which assesses the impact of individual risks on the whole reliability, is established to achieve unified integration and mapping of previous heterogeneous information, therefore providing multidimensional vulnerability representations. An empirical case study demonstrates the fault detection rate, false alarm control, and diagnostic stability of the proposed framework.

CMML2AML: Machine-learning discovery of co-mutations predictive of blast transformation in chronic myelomonocytic leukemia

Abstract Large language models (LLMs) excel at understanding natural language but struggle with explicit commonsense reasoning. A recent trend of research suggests that the combination of LLM with robust symbolic reasoning systems can overcome this problem on story-based question answering (Q&A) tasks. In this setting, existing approaches typically depend on human expertise to manually craft the symbolic component. We argue, however, that this component can also be automatically learned from examples. In this work, we introduce LLM2LAS, a hybrid system that effectively combines the natural language understanding capabilities of LLMs, the rule induction power of the learning from answer sets (LAS) system ILASP, and the formal reasoning strengths of answer set programming (ASP). LLMs are used to extract semantic structures from text, which ILASP then transforms into interpretable logic rules. These rules allow an ASP solver to perform precise and consistent reasoning, enabling correct answers to previously unseen questions. Empirical results outline the strengths and weaknesses of our automatic approach for learning and reasoning in a story-based Q&A benchmark.

Multiparametric analysis of functional group–performance relationships in tannic acid-based hydrogels from a pharmaceutical technology standpoint

Abstract. Reducing uncertainties in regional carbon balances requires a better understanding of CO2 transport in synoptic weather systems. Here, we apply the Patient Rule Induction Method (PRIM), a data-mining method to identify high-density regions for a target-class within an input parameter space, to airborne observations of potential temperature, wind speed, water vapor mixing ratio, and CO2 dry mol fraction gathered during the Atmospheric Carbon and Transport (ACT)-America Summer 2016 and Winter 2017 campaigns. ACT observations were targeted at expert-designated cases of fair weather and near-frontal warm and cold sector air at atmospheric boundary-layer, lower-, and higher free tropospheric levels (ABL, LFT, and HFT, respectively). We investigate atmospheric characteristics of these pre-defined cases and associated CO2 model-observation-differences in the mesoscale WRF-Chem model. PRIM results separate winter- and summertime observations as well as observations from ABL, LFT, and HFT with enrichment factors of 4.0–20.5 inside the PRIM box compared to the entire dataset but cannot distinguish between near-frontal warm and cold sector observations in the higher free troposphere. Analyzing of the parameter space constrained by PRIM, we find that large magnitude model observation differences preferentially associated with times when atmospheric conditions are less typical. This association suggests that PRIM could provide a useful tool for isolating atmospheric conditions with large-magnitude and non-Gaussian CO2-residuals for targeted transport model evaluation and to potentially improve inversion results during synoptically active periods.

After pointing out the problems of the conventional Rough Sets' methods that induce if-then rules hidden in a dataset called the decision table (DT), we proposed a new model for generating the DT and developed a method, statistical test rule induction method (STRIM), for inducing rules in the DT from a statistical perspective. The validity and effectiveness of STRIM were confirmed by applying it to the DT generated based on the data generation model (DGM). In general, the effectiveness of the methods inducing rules is often confirmed by applying them to classification problems that predict outputs against new inputs. From this perspective, we applied an expanded STRIM (ex-STRIM) to a classification problem and confirmed its effectiveness. This paper applies a Feedforward Neural Network and a modified ex-STRIM to the classification problem under the DGM and compares their classification performance and features as a classification method. Although the classification results strongly depend on the latent rules in the dataset and the number of learning datasets, the similarities and differences between the two methods are discussed; for example, both methods use different approaches to realize a conditional probability distribution of the output, given the input.

Vehicle electrification, automation, and sharing are reshaping urban travel demand. This study utilizes a system dynamics model to forecast travel demand in downtown Shanghai through 2035, incorporating these emerging technologies. 5,000 scenarios are evaluated using Exploratory Modelling and Analysis, and scenario discovery is conducted using the Patient Rule Induction Method. Modelling results reveal two opposite CO2 emissions trends driven by shifts in travel mode shares. (1) 433 scenarios exhibit a desirable trend, with the largest reduction to 83% of the 2022 emission by 2035; (2) Approximately half of the scenarios show a concerning trend, with the largest increase to 145% of the 2022 level by 2035. These opposing trends are attributed to variations in the pace and acceptance of autonomous vehicle adoption. Our findings underscore the necessity for strategies that enhance non-motorised transport while accelerating the development of emerging mobility technologies and encourage green travel choices among residents.

Data-driven methods such as reinforcement and imitation learning have achieved remarkable success in robot autonomy. However, their data-centric nature still hinders them from generalizing well to ever-changing environments. Moreover, labeling data for robotic tasks is often impractical and expensive. To overcome these challenges, we introduce a new self-supervised neuro-symbolic (NeSy) computational framework, imperative learning (IL), for robot autonomy, leveraging the generalization abilities of symbolic reasoning. The framework of IL consists of three primary components: a neural module, a reasoning engine, and a memory system. We formulate IL as a special bilevel optimization (BLO), which enables reciprocal learning over the three modules. This overcomes the label-intensive obstacles associated with data-driven approaches and takes advantage of symbolic reasoning concerning logical reasoning, physical principles, geometric analysis, etc. We discuss several optimization techniques for IL and verify their effectiveness in five distinct robot autonomy tasks including path planning, rule induction, optimal control, visual odometry, and multi-robot routing. Through various experiments, we show that IL can significantly enhance robot autonomy capabilities and we anticipate that it will catalyze further research across diverse domains.

Maximizing the effectiveness of Large Language Models (LLMs) requires prompt optimization, but existing approaches frequently have limited interpretability, high computational cost, and narrow generalization. We introduce RECAP, a modular, cognitively based framework for explainable and automated prompt engineering. Neurofeedback-based self-scoring, evolutionary prompt graph search, contrastive-symbolic rule induction, Pareto-based cost-accuracy optimization, an interactive debugging interface, and a shared inter-module memory layer are the six main innovations it presents. Without the need for model fine-tuning, RECAP lowers token, latency, and memory overhead while increasing prompt quality and LLM accuracy. It offers a scalable and interpretable substitute for conventional tuning pipelines and can be used in a variety of fields, including conversational AI and search.

RIONIDA: A novel algorithm for imbalanced data combining instance-based learning and rule induction

The integration of neural and symbolic systems termed neuro-symbolic AI presents a compelling path toward explainable reasoning in Artificial Intelligence (AI). While deep learning models excel at pattern recognition and generative capabilities, their opaque decision-making process has raised concerns about transparency, interpretability, and trustworthiness. This research investigates the convergence of generative AI and neuro-symbolic architectures to enhance explainable reasoning. Employing a mixed-methods methodology grounded in empirical evaluation, knowledge representation, and symbolic rule induction, the study presents a hybrid framework where large language models (LLMs) are augmented with symbolic reasoning layers, allowing for natural language generation with traceable logic paths. Experimental results on benchmark datasets such as CLEVR, e-SNLI, and RuleTakers demonstrate substantial improvements in logical coherence, reasoning accuracy, and explanation fidelity over purely neural baselines. The study further explores implications for regulated domains, including healthcare, law, and cybersecurity. This work provides a foundation for future AI systems that are powerful in generation and transparent in justification, offering an interpretable-by-design approach to responsible AI.