Search Results (452)

We study the reconciliation of several biased, partially-overlapping estimators of a latent distribution on the probability simplex (the set of non-negative credit shares summing to one) when no ground truth is available to adjudicate their disagreement—the situation faced when conversion credit is reported by independent commercial attribution sources. We cast this as non-additive information fusion and develop a four-layer fuzzy framework: a reliability-weighted fuzzy representation, an interaction-corrected conflict measure, a Choquet integral with respect to an elicited fuzzy measure (capacity) followed by an $L_1$-closure onto the simplex (a per-cell renormalization returning a valid distribution), and a Mamdani layer producing a conformally calibrated uncertainty band. Our theoretical contributions are (i) a simplex-consistency result—channel-wise Choquet aggregation with $L_1$-closure always returns a valid distribution and reduces to the weighted mean under an additive capacity; (ii) variational and axiomatic characterizations of the operator; (iii) an exact Shapley–disagreement decomposition that quantifies how non-additivity reweights redundant sources; and (iv) a Lipschitz stability bound that guarantees robustness to noisy inputs. The capacity is elicited from observable signals alone, so the method requires no labels. On a firewalled, ten-seed benchmark across eight regimes the framework attains the lowest MAE in five of eight conditions and the lowest KL in six, improving on Dempster–Shafer, correlation-clustering, and reliability-weighted baselines by 1.6–2.3% (MAE) and 3.1–4.4% (KL)—significant in four of its five targeted redundancy- and bias-dominated regimes—with the conformal step attaining nominal coverage. This advantage stems from the elicited capacity rather than the Choquet operator (a simpler ordinal integral matches it on point error); Choquet is retained for its cardinal, calibratable output. An illustrative two-source real-data study exposes a 13.7% cost-per-install distortion in the single-source view, with the (uncalibrated) band widest where the sources disagree most. Full article

Geospatial Artificial Intelligence (GeoAI) and the rapid advancement of 3D data acquisition technologies (e.g., LiDAR) have enabled scalable semantic interpretation of georeferenced large-scale 3D point clouds across different applications. 3D deep learning-based semantic segmentation on large-scale 3D point clouds typically relies on data partitioning and sampling strategies to address computational constraints, resulting in a substantial proportion of points not being directly predicted by deep neural networks. These points not directly predicted by the model require a processing step for label propagation, which is commonly handled using simple spatial proximity-based rules. While this simplification may be acceptable for some applications, it becomes critical in tasks that require precise spatial measurements and accurate object delineation, where propagation errors can directly affect downstream analyses. To bridge this research gap, this study views this process as a spatial surrogate modeling problem, where predictions from deep learning models are used to infer labels for underrepresented points based on spatial relationships. We adopt inverse distance weighting (IDW) as a transparent, deterministic, and training-free spatial post-processing strategy to examine whether explicitly incorporating spatial relationships improves segmentation outcomes. We evaluate the proposed method within an existing 3D semantic segmentation workflow for bridge inspection, a practical application requiring accurate spatial measurement and reliable object delineation. In the experiments, we use self-collected terrestrial LiDAR point clouds of bridges and associated hydraulic structures and systematically assess how neighborhood size and distance-decay parameters affect model performance on the segmentation task. Results show that this spatial post-processing step provides a modest enhancement over the conventional nearest-neighbor propagation baseline, with notable gains especially on spatially sparse or geometrically complex classes. The results also reveal class-dependent spatial effects, suggesting that different semantic classes exhibit distinct spatial dependencies during label propagation. These findings highlight the practical importance of accounting for spatial context when propagating semantic information to those points not directly predicted by deep neural networks. This is particularly important for downstream applications that require highly accurate spatial measurement and object delineation. The practical value of this post-processing step becomes more apparent in data-scarce application domains such as hydraulic-structure inspection, where large labeled point-cloud datasets and public benchmarks remain limited. In such settings, users may rely on domain-specific pre-trained models, while the original training data may not be publicly available for retraining or extensive model modification. Using a practical case study in the hydraulic domain, this study shows that deterministic post-inference label propagation can improve complete point-wise prediction from an existing model, thereby supporting the reuse of available models for domain-specific applications. The resulting response surfaces support two practical uses: site-specific calibration when limited labeled target data are available, and empirically informed initial settings, a rule of thumb, for comparable bridge-LiDAR applications when target-site labels are unavailable. Full article

Reverse engineering of printed circuit boards is still largely manual when original computer-aided design documentation is unavailable. This paper presents a semi-automatic vision-based pipeline that prepares an editable KiCad schematic draft for use in an Electronic Design Automation (EDA) workflow from paired TOP and BOTTOM board images. The method combines color-profile estimation, pad and through-hole detection, trace segmentation, optical character recognition, component inference, an explicit evidence graph and schematic export with drawn wires. A separate readability step aligns symbols to a grid and reroutes the reconstructed nets with orthogonal wires; it does not change the reconstructed netlist. The primary quantitative evaluation used twelve synthetic KiCad fixtures and three solver configurations: the default sequential pipeline, an opt-in global component solver and an opt-in probabilistic contact solver. These fixtures provide controlled regression cases and are complemented by a small exploratory acquisition trial on real photographed boards. All configurations completed all runs and passed the export round-trip validation without falling back to label-only connectivity. This round-trip check confirms consistency between the internal reconstruction and the exported schematic, but it is reported separately from electrical correctness against the KiCad reference design. The stricter reconstruction-quality criterion still failed on four stress cases involving repeated component chains, long meandering variable-width traces, circular distractors near pads and two-sided transistor layouts. The probabilistic contact solver was therefore kept as an opt-in diagnostic mode rather than enabled by default; it reduced the global pin-to-pin netlist edit distance from 642 to 525 while preserving schematic export checks. The real-board trial indicates that pad and hole detection can transfer to simple photographs, with trace extraction remaining sensitive to uncontrolled illumination and weak copper contrast. The results support the use of the system as a human-in-the-loop reconstruction assistant and identify component grouping, trace-contact reasoning, real-photograph benchmarking and safe missing-edge activation as the main remaining research problems. Full article

Direct antibody labeling is essential for immunoassays, multiplexed imaging, and biosensing; however, current methods are often time-consuming, restricted by antibody source, risk compromising protein performance, or vary with multiple steps. We introduce multiplex in situ tagging (MIST) Linker, a rapid and Fc-site-specific labeling tool that conjugates fluorophores or DNA oligonucleotides to antibodies from diverse commercial sources in as fast as 10 min using minimal starting material. MIST Linker achieves >90% cleavage upon UV exposure, facilitating rapid cyclic imaging on a single specimen. Validated across multiple species and sources, the platform outperforms conventional two-step immunofluorescence and immunohistochemistry in various tissues and cell lines. By enabling the rapid, cost-effective customization of antibody panels, MIST Linker significantly lowers the barrier to accessing antibody–DNA conjugates for spatial biology. When integrated with the spatial MIST platform and MIST-Explorer, it enables high-plex, single-cell spatial proteomics at high signal-to-noise ratios in human clinical biopsies, mouse specimens and cell lines. This toolkit provides an efficient, accessible solution for high-resolution spatial mapping, allowing for the in-depth analysis of cell subpopulations, biomarker distributions, and signaling events in complex biological specimens. Thus, MIST Linker offers a versatile, accessible, and scalable solution for antibody-labeling-based research and clinical diagnosis. Full article

Curling is a strategic sport in which shot decisions involve both expected rewards and inherent risks; expected-score distributions (ESDs)—probability distributions over possible final scores—capture this uncertainty as a risk-aware strategic indicator. Although unified frameworks predicting ESDs across all shots have been proposed, their internal behavior and learned output characteristics have not been systematically examined. We construct an all-shot ESD prediction framework grounded entirely in real-match board configurations drawn from World Curling Federation championship events and conduct a shot-wise analysis along two complementary axes: accuracy, the model’s reproduction of its training targets, and plausibility, the validity of those targets at aggregate and in-match scales. Accuracy degrades monotonically with shot number; only shot 16 admits comparison with an independent reference, while intermediate-shot labels are bootstrap rollouts of the next-shot model. An independent ground-truth probe at the first backward step (shot 15,500 boards × 95 contexts) bounds the consequence of this bootstrap structure: the resulting deviation is a small selection-induced hammer-underestimation bias common to both FiLM and concatenation chains, whose magnitude is more than an order of magnitude larger than the FiLM chain versus concatenation chain stage difference, and full multi-step verification beyond the first step is structurally out of reach. Within this scope the aggregate ESD shares the gross shape of the empirical end-score distribution and the qualitative hammer/non-hammer ordering, with a hammer-favorable offset attributed—after disentangling intent from execution on the shot-percentage 100% subset—to a label execution-noise envelope that is tighter than the realized play of top-tier matches. Within real matches the chain responds smoothly to each delivered stone and, at the directly validated terminal shot, assigns mean probability $\approx 0.71$–$0.72$ to the realized end score under intended execution (against $0.28$ for a marginal-frequency baseline), transferring from senior to junior populations. The framework is best read as an internally coherent chain anchored to a directly validated final-shot calibration, statistically plausible under intended execution with a clear boundary at execution failure. Two structural limitations remain: rare high-magnitude outcomes are scarce in real data and produce a heavy upper tail of accuracy errors, and a single fixed execution-noise envelope that is tighter than top-tier realized play accounts for the aggregate hammer-side offset and motivates recalibration against real-match execution statistics. Full article

Background/Objectives: Consumption of dairy products and plant-based dairy alternatives (PBDAs) can be examined within broader configurations of food-related orientations rather than as isolated product choices. This study aimed to identify food-related consumer segments based on food involvement, attention to on-pack product information, and sustainability-related food-choice orientations, and to characterise these segments in relation to reported consumption frequencies of dairy products, PBDAs, and meat, fish and legume dishes, as well as health-oriented food-choice criteria. Methods: A cross-sectional survey was conducted among 1508 Polish adults responsible or co-responsible for household food purchasing. Principal component analysis was used to identify underlying food-related dimensions, and the retained component scores were entered into a two-step cluster analysis. Differences between clusters were examined using chi-square tests and one-way ANOVA. Results: Six dimensions were retained: sustainable and ethical choices, meat reduction, food involvement, product-information importance, shopping-list use and food-waste avoidance. Five clusters were identified, reflecting distinct configurations of these dimensions. PBDA and legume-dish consumption were most frequent in the sustainability and meat-reduction-oriented cluster, although dairy products and meat remained part of the reported diet. High food involvement and label/quality attention co-occurred with a more conventional consumption pattern, whereas PBDA and legume-dish consumption were lowest in more conventional and lower-sustainability clusters. The low-engagement cluster showed a more selective pattern of PBDA and legume-dish consumption. Conclusions: This study identified five food-related consumer segments and showed that reported PBDA consumption was embedded in heterogeneous dietary patterns rather than functioning as a simple substitute for dairy products. These findings indicate that reported PBDA consumption is segment-dependent and cannot be assumed to reflect reduced dairy consumption or a consistently sustainability- or health-oriented dietary pattern. Full article

The central charge occupies a unique role in conformal field theory, simultaneously serving as a measure of degrees of freedom, as the determinant of Casimir energy through modular transformations, and as an obstruction to the naive extension of the Witt algebra. The Virasoro central extension itself is rigid: it fixes c as a label of a given conformal field theory. In this work, we propose that higher categorical coherence—the pentagon and hexagon constraints governing fusion and braiding data, one level above the cocycle responsible for the Virasoro extension—supplies an additional, physically controllable handle. We show that controlled deformations of this higher coherence (higher categorical coherence breakdown, HCCB), implemented consistently through anomaly inflow, shift the effective central charge read out by anomaly-sensitive observables in quantized steps, opening the possibility of treating the measured central charge not as a fixed label but as an experimentally addressable piecewise-quantized quantity. We then focus on the fractional quantum Hall effect (FQHE), where the chiral central charge $c_{-}$ directly governs the quantized thermal Hall conductance. After reviewing the role of edge conformal field theories and current bounds on thermal transport, we propose experimental modifications—such as engineering multi-component edge states, coupling to non-Abelian quasiparticles, or introducing controlled categorical perturbations—that could render higher coherence breakdown detectable as shifts in the effective central charge. Two further elements complete the program. First, we show that within the consistent framework, all route- and bracketing-dependent observables vanish identically (route blindness), so that the pentagon and hexagon interferometers and thermal Y-junction networks we design operate as precision null tests of the modular-functor axioms themselves—the axioms stating that anyonic amplitudes are determined by the topology of a process rather than by the bookkeeping route used to compose it. Second, we show that a quantized remnant of route sensitivity survives in exactly one consistent form: the holonomy of closed cycles of categorical controls, realizing a central-charge pump for which the integer count per cycle is a family invariant beyond any static stacking description. The resulting framework provides both a conceptual reinterpretation of the central charge as a higher obstruction in categorical terms and a concrete experimental route for probing its dynamical behavior. Beyond the quantum Hall setting, these ideas suggest a broader program: anomalies, topological phases, and even string worldsheet central charges may admit reinterpretation through higher coherence. We conclude by outlining a research agenda in which categorical methods yield new experimental observables, potentially transforming the interplay between mathematics, condensed matter physics, and high-energy theory. Full article

Marine pollution monitoring increasingly relies on Biological Early Warning Systems (BEWSs), which use living organisms as continuous, integrative sentinels of water quality. The moon jellyfish Aurelia aurita is a sensitive but under-exploited candidate for this role. We present a computer-vision BEWS pipeline that is unsupervised at inference time and operates without labelled pollution-response data, converting side-view aquarium video of single A. aurita medusae into a binary pollution alarm. Per-frame YOLO bounding-box detections are reduced to a continuous bell-area signal and a centroid trajectory, from which eleven pulsation, kinematic, and detection-quality features are extracted on 60 s sliding windows. A per-animal baseline is fitted on a clean-water baseline (recommended ≥15 min), and a two-layer detector—fast outlier detection on the mean absolute z-score with a k-of-N rule, plus one-sided CUSUM (cumulative sum) accumulation—flags any sustained deviation. Validation on six adult medusae exposed to diesel-WAF detected all six animals (95% CI 54–100%) and produced no false alarms in 203 clean-window opportunities (exact 95% upper bound 1.8%; rule-of-three estimate ≈1.5%). First-alarm latencies ranged from 1.0 to 23.7 min, and the observed responses were described as three descriptive patterns in this pilot dataset: sharp step-change, slow drift, and mixed. The deployed anomaly scoring step contains no neural-network weights, runs in under 300 lines of Python, and is designed for field-portable use in settings where a stationary side-view camera can be positioned alongside an aquarium, although field validation remains required. Per-animal anomaly detection accommodates the strong inter-individual variability of the diesel-WAF response that limits supervised clean-versus-polluted classification at this sample size. Full article

Accurate annotation of secreted and membrane proteins requires detecting N-terminal secretion signals, locating their cleavage sites, and distinguishing secretion-signal classes, while also predicting full transmembrane topology for both alpha-helical and beta-barrel proteins. Current tools typically address either whole-protein topology with a generic signal-peptide category or signal-peptide type classification without integrated topology annotation, leaving end-to-end labels incomplete when both features must be resolved together. Here, we present PRED-TMSdeep, a deep learning method that jointly predicts transmembrane topology and three signal peptide classes: secretory pathway/signal peptidase I (Sec/SPI), secretory pathway/signal peptidase II (Sec/SPII), and twin-arginine translocation/signal peptidase I (Tat/SPI). We introduce a two-step constrained decoding procedure that first detects transmembrane segments and signal peptides and then resolves global orientation and refines boundaries under stricter biological constraints. On redundancy-reduced datasets curated from the Orientation of Proteins in Membranes and the Protein Data Bank of Transmembrane Proteins, PRED-TMSdeep matches leading predictors for segment-level topology while improving signal peptide classification and yielding the highest overall top-1 cleavage-site accuracy. Top-1 cleavage-site accuracy reached 89.2%, compared with 84.7% for TMbed and 86.2% for SignalP 6.0, mainly reflecting strong performance on the predominant Sec/SPI class. The software is available as a web server and a batch command-line tool with pretrained models and reproducible workflows. Full article

Large language models and vision–language models already achieve strong results on reasoning tasks, but their reliability under controlled assessment-style conditions remains insufficiently characterized. This paper presents a benchmark study of multimodal multiple-choice mathematical reasoning using 324 Austrian Mathematical Kangaroo competition problems (2022–2024), including both text-only and diagram-dependent items. We evaluate five state-of-the-art models under a controlled protocol that isolates two factors: input modality and prompt format. We compare a strict short-answer condition requiring a single option label (one_liner) with a structured condition eliciting step-by-step reasoning and an explicit final answer (full) while enforcing deterministic decoding and rule-based answer extraction. Performance is assessed using accuracy, abstention rates, and contest-style scoring, supported by paired and unpaired statistical analyses and a structured error taxonomy. The results show that prompt format is the primary driver of performance: structured prompting yields substantial gains across all the models, particularly on text-only items. In contrast, visual-text problems remain consistently harder, with a robust performance gap that persists across prompting conditions, indicating persistent limitations in visual grounding. Model comparisons are additionally influenced by response strategies, especially abstention behavior under strict output constraints. An error analysis reveals systematic failure modes, including constraint violations, inappropriate strategy selection, and diagram misinterpretation, alongside structured biases in multiple-choice selection under constrained prompting. Overall, the findings demonstrate that measured performance is highly sensitive to the interaction between prompt format and input modality. This underscores the importance of treating prompting, decoding, and answer extraction as integral components of evaluation in assessment-oriented settings, where reliability and reproducibility are central. Full article

Ovarian cancer (OC) is typically treated with cytoreductive surgery (CRS). Hyperspectral imaging (HSI) is an emerging non-invasive, label-free technique that enables whole-area scanning, making it a promising tool for real-time tumour recognition. However, developing tumour recognition algorithms requires a foundational understanding of spectral variability in normal tissues. This study focusses on the in vivo spectral profiles of key abdominal and pelvic organs encountered during CRS, including the uterus, ovaries, intestines, mesentery, omentum, peritoneum, and fallopian tubes, and evaluates the potential for organ recognition using HSI data. Intraoperative HSI data were from healthy patients. Two machine learning models, a support vector machine (SVM) and a 3D convolutional neural network (3DCNN), were trained to classify the organs based on their spectral signatures. In total, 15 patients were included in the dataset. The 3DCNN slightly outperformed the SVM in terms of the average accuracy (0.889 vs. 0.878), sensitivity (0.648 vs. 0.604), specificity (0.936 vs. 0.930), and Dice Similarity Coefficient (0.595 vs. 0.569). This study demonstrates the feasibility of using HSI for organ differentiation in the clinical setting, although in some cases separability remains a challenge, especially when organs have similar spectra. This is a critical step towards a generalizable in vivo abdominal tumour recognition algorithm, by carefully investigating spectral fingerprints of abdominal tissues. Full article

Crash severity prediction is critical for emergency response, infrastructure spending, and risk communication. Although machine learning has been widely applied to this problem, three gaps prevent practical deployment: uncalibrated probability scores, SHAP-based explanations whose faithfulness has not been verified, and models never tested outside their training jurisdiction. The proposed framework, SAE-XCrash (Safety-Aware and Explainable Crash Severity Prediction), addresses all three using two public datasets—US-Accidents (7.0 million records, 2016–2023) and UK STATS19 (approximately 1,010,000 records, 2016–2022)—with strict temporal splits throughout. Notably, the US-Accidents severity label measures traffic disruption duration, not injury outcome; results should be interpreted accordingly. Previously unknown label-schema drift led to a revised binary target with Severity 4 as the only positive class. Five classifiers are compared. Post hoc isotonic calibration reduces Expected Calibration Error by 97.3% at negligible discrimination cost. A four-step quantitative SHAP audit confirms statistically significant deletion faithfulness; however, explanation stability fails at realistic perturbation levels (54.3% low-stability fraction at sigma = 0.05), driven by spatial data sparsity in sparse geohash cells—a negative result that carries direct operational implications for deployment. A three-tier cross-dataset transfer experiment (zero-shot, recalibration, full retrain) shows that temporal features transfer robustly across jurisdictions, while spatial memorization is the primary generalization barrier. All code, split indices, and model artifacts are publicly available. Full article

Breast cancer is a heterogeneous disease characterized by substantial molecular diversity and variable treatment outcomes across patients. Despite advances in targeted and systemic therapies, anticipating individual benefit remains a major clinical challenge. In this context, Artificial Intelligence (AI) can support precision oncology by integrating high-dimensional molecular profiles with clinical and pharmacological information. Here, we present B.R.E.A.S.T. (Breast canceR Enhanced AI-Supported Therapy), an interpretable machine learning framework designed to predict therapy outcome from tumor proteomic profiles integrated with clinical and treatment annotations. Proteomic data from The Cancer Genome Atlas (TCGA) and The Cancer Proteome Atlas (TCPA) were harmonized with outcome and therapy information, and thirteen supervised classifiers were systematically evaluated using stratified 5-fold cross-validation. Therapeutic outcome labels were operationally defined by integrating available treatment response annotations with complementary clinical outcome information. Across both cohorts, ensemble-based models consistently achieved the most stable and highest discriminative performance, supported by learning-curve analyses and consistent behavior across independent datasets. To enhance interpretability, we implemented a two-step feature selection strategy combining model-specific importance measures with a global consensus ranking, enabling the identification of a compact set of robust proteomic biomarkers associated with therapeutic outcome. Top-ranked features mapped to molecular programs relevant to breast cancer progression and treatment sensitivity, including regulators of cell survival, DNA damage response, PI3K/AKT/mTOR signaling, and invasion-related processes. Re-evaluation using only the top 30 globally ranked features preserved high predictive performance across both independent breast cancer cohorts, indicating that a parsimonious proteomic signature captures core molecular determinants of outcome. Overall, B.R.E.A.S.T. provides a robust and generalizable proteomics-driven framework for modeling outcome-associated therapeutic response patterns and supporting biologically informed biomarker discovery in breast cancer. Full article

Automatic dental age estimation in deep learning applications relies on accurate assessment of tooth development stage (TDS) and regions of interest (ROIs). Since dental development is population-specific, a recent study showed that a two-step approach using an ethnicity-specific reference dataset can improve explainability, transferability across populations, and agreement with chronological age. Therefore, this study aims to investigate whether the proposed population-aware framework can be extended toward greater automation by using Faster Region-Based Convolutional Network (R-CNN) for third-molar detection. A total of 1110 digital panoramic radiographs and 1980 labeled ROIs were used to train and validate four pretrained backbones: AlexNet, DenseNet-201, GoogLeNet, and VGG-16. DenseNet-201 achieved the best training performance and overall validation performance (accuracy = 95% and marco recall = 0.75), followed by GoogLeNet (84.7% and 0.80), VGG-16 (86.6% and 0.75), and AlexNet (79.8% and 0.58). Validation results showed relatively high recall across most TDS stages, whereas precision was lower because of false-positive (FP) detections. Performance was only minimally affected by the intersection-over-union (IoU) threshold, suggesting stable localization behavior. The trained Faster R-CNN models provide a clinical-oriented assistive tool for automatic third-molar detection and TDS classification, serving as an intermediate component that can potentially support more efficient ethnicity-specific dental age estimation in future clinical practice, forensic investigations, and pediatric dentistry. Full article

Suffering from the well-known Hughes phenomenon, hyperspectral image (HSI) classification is still very challenging, which is mainly due to the high-dimensional features of HSIs and relatively limited training samples. To systematically represent the spatial and spectral relationships among the HSI data, a connection center evolution (CCE) and local similarity-based data gravitation integrated classification (CCE-LSDGC) model is proposed for the classification of HSIs. In the first step, the cosine similarity matrix of the labeled samples and their neighboring pixels is integrated with the CCE theory to enlarge the size of the training set. Then, the cosine similarity matrix of the test pixel and its neighboring pixels is used to define their local spectral and spatial similarity. The similarity is taken as the local mass of neighbors that weight the contribution of different neighbors in a data gravitation model. This effectively alleviates the interference of local heterogeneous pixels and noise. Finally, each test pixel is labeled to the class whose training samples exerted the largest average data gravitation in the local joint region. Comprehensive experiments conducted on two benchmark and two real-world HSIs datasets have verified the superiority of the CCE-LSDGC model compared to a few state-of-the-art deep learning methods. In particular, the proposed method shows high performance on the HSIs with limited training samples. Full article