Predicting patient mortality in the intensive care unit (ICU) is critical for timely and effective clinical decisionmaking. While deep learning models have been increasingly applied to Electronic Health Records (EHR) data, modeling heterogeneous and often incomplete clinical data remains a major challenge. Existing approaches frequently fail to fully capture both modality-specific characteristics and inter-modal correlations across diverse data types. In this Paper, we propose MVAE4EHR, a multimodal variational autoencoder framework that jointly learns modality-specific and shared intermodality representations from heterogeneous EHR data. The model maintains the specificity of individual modalities (e.g., time-series vitals, lab results, static demographics, and clinical notes) while capturing their correlations through a unified latent representation. This dual-objective design enables our framework to produce rich, disentangled, and high-quality representations beneficial for downstream tasks. We evaluate our method on the publicly available MIMIC-III dataset. Our model outperforms unimodal baselines and other multimodal fusion models in mortality prediction, achieving an AUROC of 0.90 and AUPRC of 0.65. Our findings indicate that modeling inter-modality correlations and modality-specific features enhances the predictive performance of mortality prediction models. This method offers a promising path for creating interpretable and effective clinical decision-support tools.

In the present paper, resolution-based automated reasoning theory and algorithm in a finite chain lattice-valued proposition logic are focused. Concretely, the resolution principle, which is based on a finite chain lattice-valued propositional logic FCLP(X) is investigated. And soundness theorem and completeness theorem of this resolution principle are also proved. In order to realize resolution, the concrete algorithm of resolution is discussed. It is hoped that this research will make forward theoretical research of automated reasoning based on lattice-valued logic.

In the workflow management system, the process definition is finished by the process definition tools and explained and enacted by the workflow enactment service. The method that the application logic and the process logic are separated improves the reuse rate of the software and the efficiency of the system by modifying the process model rather than the material functions. We propose an automated reasoning method to validate the process definition in this paper. The method includes two steps. In the first step, the typical relationship between the activities is discussed, and then the transformation rules from the relevant activities to a clause set is given; the automated reasoning based on path searching is proposed for judging the satisfiability of the clause set and finding out the validation of the process definition in the second step. Experimental results show that this method is effective for WfMS.

In this paper, a multilayered fuzzy clustering method based on distance and density (MFCDD) is proposed. The first layer's algorithm deals with the original data points, the upper with the cluster centers of the contiguous lower layer. In each layer it identifies the cluster number automatically. It calculates the density and density set of each data point based on distance matrix; then chooses one data point randomly and judges whether every element in the selected data point's density set is in the same cluster with itself, this process is repeated till all data points have been selected. In order to find the optimum value of the parameters, we adopt an objective function using entropy on the upmost layer. Clustering analysis of MFCDD has been performed and the experimental results show that a high recognition rate can be achieved.

Fuzzy clustering is capable of finding vague boundaries, but its time complexity is usually high, and the need to specify complicated parameters hinders its use. Here, a new fuzzy clustering method based on distance and density (FCDD) is proposed, which automatically identifies the cluster number. It calculates the density and density set of each data point and selects any data point at the beginning of the FCDD algorithm. Next it judges whether every element in the chosen data point's density set is in the same cluster with itself. This process is repeated until all data points has been chosen. The method does not require finding the cluster center and density values are calculated only once. It requires two parameters that are easy to specify and is able to find the natural clusters in the data. In order to find the optimum values of the parameters with respect to the specified number of the cluster, we construct a target function using entropy. Cluster analysis of the two methods has been performed on several data sets and the experimental results show that a high recognition rate can be achieved.