Ion Technology Center (Japan)

Rule representations significantly influence the search capabilities and decision boundaries within the search space of Learning Classifier Systems (LCSs). However, it is very difficult to choose an appropriate rule representation for each problem. Additionally, some problems benefit from using different representations for different subspaces within the input space. Thus, an adaptive mechanism is needed to choose an appropriate rule representation for each rule in LCSs. This article introduces a flexible rule representation using a four-parameter beta distribution and integrates it into a fuzzy-style LCS. The four-parameter beta distribution can form various function shapes, and this flexibility enables our LCS to automatically select appropriate representations for different subspaces. Our rule representation can represent crisp/fuzzy decision boundaries in various boundary shapes, such as rectangles and bells, by controlling four parameters, compared to the standard representations such as trapezoidal ones. Leveraging this flexibility, our LCS is designed to adapt the appropriate rule representation for each subspace. Moreover, our LCS has a generalization bias to produce as many crisp rules as possible. Experimental results on real-world classification tasks show that our LCS significantly outperformed LCSs with popular rule representations in test classification accuracy on up to 17 of the 25 datasets tested.

To address the dilemma of limited accuracy in traditional physics-based hydrological models and poor generalization in pure data-driven models, this study proposes a novel hybrid hydrological model (XAJ-BiGRU) based on the concept of "modular deep replacement." This model uses the Xinanjiang model as its physical framework, employing a Bidirectional Gated Recurrent Unit (BiGRU) network to precisely replace the original model's error-prone evapotranspiration and runoff generation modules. Meanwhile, it retains physically meaningful modules, such as routing, as constraints, thereby constructing a "grey-box" model that balances high accuracy with physical interpretability. In a case study validation in the Niulanjiang River Basin, the proposed XAJ-BiGRU model demonstrated significant advantages in both runoff simulation accuracy and stability compared to the traditional Xinanjiang model and a pure data-driven model, effectively overcoming the structural deficiencies of the former and the overfitting issue of the latter. The research indicates that this deep integration strategy successfully combines the reliability of physical mechanisms with the flexibility of data-driven approaches, offering an effective pathway for developing next-generation intelligent hydrological forecasting models.

This paper proposes a new paradigm of the EmbodiedCyber-Physical System (Embodied CPS, ECPS) to address theissues of the disconnection between physical laws andintelligent decision-making and the insufficientinteraction with dynamic environments in the modeling andsimulation of traditional CPS. ECPS achieves unifiedmodeling of physical laws and autonomous decision-makingthrough the "perception-decision-action" closed loop.Toverify ECPS, an embodied space framework based onModelica/MWorks is designed. Through three majortechnological innovations: constructing an embodied domainmodeling specification and embedding the Navier-Stokesequations into the training of the policy network;expanding the syntax and semantics of Modelica,encapsulating physical constraint reinforcement learningcomponents, and establishing a gradient interactionprotocol between the Physics-Informed Neural Network (PINN)and Modelica equations; building a digitaltwin-hardware-in-the-loop co-simulation platform based onthe FMI/SSP protocol to establish a collaborativeverification link between high-precision physicalsimulation and real-time decision-making.Taking theUnmanned Surface Vehicle (USV) as the carrier, thefull-process method from dynamic modeling, reinforcementlearning strategy training to virtual-real environmentco-simulation is demonstrated. Experiments verify theeffectiveness of this framework in achieving theclosed-loop coupling of physical simulation and intelligentdecision-making under complex sea conditions, providing amethodological foundation for interpretable modeling andverifiable simulation in the development of embodiedintelligence.

Background/Objective: Frailty is a growing public health challenge in Japan's rapidly aging population, where 28.8% are aged ≥ 65. While multicomponent interventions have shown potential in preventing frailty, traditional face-to-face programs face accessibility challenges. This study evaluated the feasibility and short-term changes of an online multicomponent intervention on frailty-related factors among community-dwelling older adults. Methods: In this single-arm feasibility study, 132 participants (mean age 75.7; standard deviation 4.8 years, 65.2% women) completed a six-week online intervention combining exercise, nutrition education, cognitive activities, and the Coimagination Method to foster social connections, meeting weekly for 75 min sessions in groups of up to 10 participants. Results: The intervention demonstrated feasibility with a 96.4% retention rate and a 94.0% average participation rate. While no significant changes were observed in physical activity levels, autonomic nervous system indicators, or cognitive function, carotenoid scores and hemoglobin concentration improved significantly, with more pronounced improvements among women than men. Conclusions: This study demonstrates the high feasibility of online multicomponent interventions for older adults and suggests potential benefits for nutritional status markers, particularly among women. These findings indicate a promising and accessible approach to frailty prevention, though randomized controlled trials with longer intervention periods and direct frailty assessments are required to establish effectiveness conclusively. Study Trial registration: UMIN Clinical Trials Registry (UMIN000053089).

It is important but also challenging to control the full shape of the molecular weight distribution in polymerization processes, since it is an infinite dimensional probability density function (PDF). In this work, a self-optimizing control (SOC) strategy is adopted to achieve the aim of PDF shaping by maintaining some elaborately selected controlled variables (CVs) at constant set points through online feedback control, even in the presence of uncertainties. To find optimal CVs, finite moments rather than the full shape, which corresponds to an infinite-dimensional space, of the PDF are adopted as elements to parametrize CVs, while the optimization problem is to minimize the distance between the actual PDF and the target PDF. The proposed SOC-PDF method is demonstrated more effective than the existing stochastic distribution control method through a pilot semibatch styrene polymerization case study.

INTRODUCTION: Focal point tracking and temperature monitoring are required for the safe and effective performance of magnetic resonance-guided high-intensity focused ultrasound (MRgHIFU) surgery on organs with high respiratory motion such as the liver. It is also important to shorten the acquisition time of MR images used for tracking and temperature measurement. We have focused on Compressed Sensing (CS) technique [1] that reconstructs MR images from short-time measurement data using sparsity, and proposed an MR temperature thermometry method using this technique In this extended abstract, a focal point prediction method using MR images acquired using CS technique and pre-operative MR images is proposed.

In this study, we propose a robust method for automotive tracking using high-speed stereo vision. The tracking is achieved using a shared correlation filter for stereo images that can be captured at high speed using the features of high-framerate images. In addition, by adjusting the tracking window for each automobile, robust tracking can be performed even in the case of partial occlusion with respect to visually overlapping automobiles. Experiments on in-vehicle camera images were performed at a frame rate of 500 fps to validate the efficacy of the proposed approach.

Mineral nutrition of orchard soil is critical for the growth of fruit trees and improvement of fruit quality. In the present study, the effects of soil mineral nutrients on peach fruit quality were studied by using artificial neural network model. The results showed that the four established ANN models had the highest prediction accuracy (R 2 = .9735, .9607, .9036, and .9440, respectively). The results of prediction model sensitivity analysis showed that available B, Ca, N, and K in the soil had the greatest influence on the single fruit weight, available Fe, K, B, and Ca in the soil had the greatest effect on fruit soluble solid content, available Ca, N, B, and K in the soil had the greatest influence on the fruit titratable acid content, and available Ca, Fe, N, and Mn in the soil had the greatest effect on fruit edible rate. The response surface methodology analysis determined the optimal range of these mineral elements, which is critical for guiding precision fertilization in peach orchards and improving peach fruit quality.