To construct the micro collision avoidance intelligent unit, efficient solutions have been pursued by modeling a specific collision-sensitive neuron, the lobula giant movement detector (LGMD), identified in the locust’s visual neural system. However, the existing LGMD models cooperating with a threshold-based collision avoidance strategy are less robust in the presence of visual variability, including fluctuations in contrast and noise. Intriguingly, biological research underscores the potential of the single-peak modeling through the η function, suggesting a non-threshold collision avoidance strategy by identifying the peak response moment. Adapting the biological finding to the engineering applications, we establish a novel modeling method based on η-form perceptron, making the response curve behave like the η function. The single peak within the η-function-form response occurs before the collision, offering reliable collision warning cues. Numerical experiments substantiate the effectiveness of the proposed model against visual variability. This study extends the limits of the applications of LGMD models in diverse scenarios, and explores the potential of modeling neural computation in constructing micro-intelligent units.
This study investigated the stable nanocomplexes between Zein (a hydrophobic plant protein), and chitosan (CS) via electrostatic interactions, followed by the co-delivery of different polyphenols, including curcumin, quercetin, and rutin, together with Lactobacillus casei, to construct a dual-encapsulated Pickering emulsion (DEPE) within a layered emulsion carrier. The formation of DEPE was found to depend on CS concentration, pH, and the synergistic effects of polyphenols and probiotics. The results showed that the stability of DEPE was influenced by CS concentration and pH, with optimal performance achieved at pH 3.0 and 0.5 mg/mL CS, yielding high encapsulation efficiencies (EE) for both oil (89.73%) and probiotics (87.76%). The hierarchical interfacial structure effectively protected probiotics in acidic environments and significantly enhanced emulsion clarity, deformation resistance, and thermal stability. Moreover, DEPE improved the survival of Lactobacillus casei during in vitro gastric digestion and storage. In conclusion, this study provides a theoretical basis for the practical application of probiotic-loaded emulsions in the development of innovative functional foods.
Accurate uncertainty quantification in structural engineering often requires estimating the full response distribution including both the cumulative distribution function (CDF) and the complementary cumulative distribution function (CCDF), rather than a single failure probability. This paper proposes a novel active learning method guided by the probability of misclassification for Gaussian process (GP) modeling to efficiently estimate these distributions for computationally expensive models. The locally defined probability of misclassification, originally meaningful only for a fixed threshold, is generalized to a global version tailored for multi-threshold estimation of CDF and CCDF. This provides a threshold-wise upper bound on the discrepancy between the posterior-predictive estimator and its mean-based approximation under the GP posterior. A tail-weighted stopping criterion placing greater emphasis on low-probability regions is then introduced to control the overall residual error in distribution estimation. Reformulating this criterion leads to a one-step learning function defined entirely in the input space, allowing all thresholds of interest to be considered simultaneously in each learning step and promoting globally efficient sampling. Numerical experiments demonstrate that the proposed method accurately recovers reference CDF and CCDF with fewer true-model evaluations than existing representative distribution-oriented active learning approaches.
In this paper, we propose CorrFuse, a unified cross-modal dynamic fusion framework for multimodal sentiment analysis (MSA). CorrFuse is designed to address several common challenges in MSA. These include modal asynchrony, signal degradation, limited static fusion strategies, and inconsistent ordinal predictions. First, we design a Cross-modal multi-round Embedding (CE) module. This module iteratively integrates visual and acoustic information using a multi-round cross-attention mechanism. It effectively mitigates the problems of modal asynchrony and signal degradation. Second, we introduce a sentiment-intensity-conditioned dynamic attention mechanism and a correlation alignment loss. These allow real-time adaptive adjustments of fusion weights and dynamically suppress noise interference. Finally, we adopt the Consistent Rank Logits (CORAL) ordinal regression. This is combined with a multi-task collaborative optimization strategy. The discrete and continuous dual constraints effectively guarantee monotonicity and semantic continuity in sentiment prediction. Experimental results on CMU-MOSI, CMU-MOSEI, CH-SIMS and CH-SIMS v2 datasets demonstrate that CorrFuse consistently outperforms existing methods. It shows strong performance and potential practical applicability.
Karst-influenced deep excavations involve coupled geological, hydraulic, structural, and observational uncertainties that cannot be represented adequately by deterministic checks alone. This study develops a monitoring-updated reliability framework for the Baiyun Dongping Station excavation in Guangzhou, China. Project documents, karst investigation records, a documented water-inrush event, field monitoring data, APDL-derived geometry, and direct global–local coupled simulations are integrated through a source-path-response interpretation. Formal probability statements are restricted to two traceable criteria: a 24 mm wall-displacement warning threshold and a 270 m3/h severe-flow threshold, while support force, groundwater change, and adjacent-asset response are treated as diagnostic or validation evidence. Bayesian updating is used to reduce uncertainty in physically meaningful parameters, including rock stiffness, cavity-roof tensile strength, fracture conductance, and effective hydraulic head. Posterior standard deviations decrease by 38.5–81.8 %. Direct simulations give severe-flow exceedance probabilities of 0.0407, 0.1386, and 0.0631 at Stages 4, 5, and 6, respectively. Effective head and fracture conductance dominate post-connection flow, whereas roof strength controls pathway initiation and soil-rock stiffness controls stress transfer. External predictive coverage is 0.70 for wall response but only 0.20 for held-out settlement, indicating that the framework supports hydraulic reconstruction and wall-warning assessment, but not a general adjacent-asset reliability claims under the available field evidence.
