Search

Cardiac digital twins provide a physics and physiology informed framework to deliver personalized medicine. However, high-fidelity multi-scale cardiac models remain a barrier to adoption due to their extensive computational costs. Artificial Intelligence-based methods can make the creation of fast and accurate whole-heart digital twins feasible. We use Latent Neural Ordinary Differential Equations (LNODEs) to learn the pressure-volume dynamics of a heart failure patient. Our surrogate model is trained from 400 simulations while accounting for 43 parameters describing cell-to-organ cardiac electromechanics and cardiovascular hemodynamics. LNODEs provide a compact representation of the 3D-0D model in a latent space by means of an Artificial Neural Network that retains only 3 hidden layers with 13 neurons per layer and allows for numerical simulations of cardiac function on a single processor. We employ LNODEs to perform global sensitivity analysis and parameter estimation with uncertainty quantification in 3 hours of computations, still on a single processor. [ABSTRACT FROM AUTHOR]

This paper studies a class of asymptotically almost periodic recurrent neural networks involving mixed delays. By utilizing differential inequality analysis, some novel assertions are gained to validate the asymptotically almost periodicity of the addressed model, which generalizes and refines some recent literature works. In the end, an example with its numerical simulations is carried out to validate the analytical results. [ABSTRACT FROM AUTHOR]

The work presented in this paper is an investigation of the prediction of amplitudes of the specific grinding force components. An improved method for artificial neural networks (ANNs) establishment is proposed here allowing accurate prediction of specific normal and tangential grinding forces. This method can determine the optimal set of inputs to be used for these ANN. This set of inputs is composed of significant factors and interactions among factors that could possibility offer the best learning and generalization of ANNs simultaneously. A 48-run experimental design (MED) is used in this research to train the ANNs and a total of 81 experiments are conducted to test the generalization performances of ANNs. Results have indicated that the developed ANNs show low deviations from the training data, and acceptable deviations from the testing data. In addition, the accuracies of these ANNs are found to be significantly better than those of other approaches used for modelling of the specific grinding force components. These approaches use regression models, power models, genetic algorithms or the common ANNs for which only factors of the MED are usually used in the input layer. [ABSTRACT FROM AUTHOR]

A neural network model for solving an assortment problem found in the iron and steel industry is discussed in this paper. The problem arises in the yard where steel plate is cut into rectangular pieces. The neural network model can be categorized as a Hopfield model, but the model is expanded to handle inequality constraints. The idea of a penalty function is used. A large penalty is applied to the network if a constraint is not satisfied. The weights are updated based on the penalty values. A special term is added to the energy function of the network to guarantee the convergence of the neural network which has this feature. The performance of the neural network was evaluated by comparison with an existing expert system. The results showed that the neural network has the potential to identify in a short time near-optimal solutions to the assortment problem. The neural network is used as the core of a system for dealing with the assortment problem. In building the neural networks system for practical use, there were many implementation issues. Some of them are presented here, and the fundamental ideas are explained. The performance of the neural network system is compared to that of the expert system and evaluated from the practical viewpoint. The results show that the neural network system is useful in handling the assortment problem. [ABSTRACT FROM AUTHOR]

In this paper, we introduce a new kind of approximate weakly efficient solutions to the set-valued vector equilibrium problems with constraints in locally convex Hausdorff topological vector spaces; then we discuss a relationship between the weakly efficient solutions and approximate weakly efficient solutions. Under the assumption of near cone-subconvexlikeness, by using the separation theorem for convex sets we establish Kuhn-Tucker-type and Lagrange-type optimality conditions for set-valued vector equilibrium problems, respectively. [ABSTRACT FROM AUTHOR]

Abstract  We propose a novel approach that aims to realize autonomous developmental intelligence called Intelligence Dynamics. We emphasize two technical features of dynamics and embodiment in comparison with the symbolic approach of the conventional Artificial Intelligence. The essential conceptual idea of this approach is that an embodied agent interacts with the real world to learn and develop its intelligence as attractors of the dynamic interaction. We develop two computational models, one is for self-organizing multi-attractors, and the other provides a motivational system for open-ended learning agents. The former model is realized by recurrent neural networks with a small humanoid body in the real world, and the later is realized by hierarchical support vector machines with inverted pendulum agents in a virtual world. Although they are preliminary experiments, they take important first steps towards demonstrating the feasibility and value of open-ended learning agents with the concept of Intelligence Dynamics. [ABSTRACT FROM AUTHOR]