As Mashup technology advances, the number of Web APIs available continues to increase year by year. As a result, identifying and selecting suitable Web APIs from the wide array of options remains a challenging issue. To address this issue, interactive Web API recommendation has been introduced to simplify the process of Web API selection, assisting users and developers in invoking APIs to meet their business or software development needs. Although there have been some collaborative filtering-based methods for Web API recommendation, the representation of Mashups and APIs is not accurate enough due to the sparsity of Mashup-API interactions. As a result, their performance is often limited. To enhance the recommendation performance, we propose an interactive Web API recommendation approach based on composition-supervised LightGCL, named CS-LightGCL. CS-LightGCL derives Mashup and Web API embeddings by utilizing a simple yet effective graph contrastive learning paradigm LightGCL in which composition-supervised learning is designed and included to guide the embedding process simultaneously. Extensive experiments on a real-world dataset demonstrate that CS-LightGCL outperforms the baseline methods. The code is available at: https://github.com/IntelligentServiceLab/CS-LightGCL.

Silicon carbide (S iC) MOSFETs have better dynamic characteristics and higher operating temperatures, and are more suitable for high temperature, high-frequency and high-voltage applications due to their fast turn-on and turn-off speeds. Since high-voltage SiC MOSFET devices can effectively reduce circuit system complexity, the application advantages of high-voltage SiC MOSFET devices are highlighted. The lack of fast and accurate simulation models is one of the key problems facing the current development of high-voltage SiC MOSFET device applications. This paper provides a behavior modeling of high-voltage SiC MOSFETs considering the effect of variable temperature. The model further characterizes and fits temperature-sensitive parameters based on the characteristics of the behavior model parameters for the differences in static characteristics of high-voltage SiC MOSFETs under different temperature conditions, and the model optimizes the SiC MOSFET equivalent circuit and considers the internal parasitic parameters of SiC MOSFETs to improve the accuracy of the model fitting. In this paper, the behavior modeling of a 6.5kV/400A SiC MOSFET device developed by State Grid Smart Grid Research Institute Co., Ltd. and verifies the validity and accuracy of the model at different temperatures by comparing the simulation data of the static model at different temperatures with the test results. Furthermore, the accuracy and convergence of the model dynamic characteristics are verified by a double-pulse test circuit and LLC resonance simulation circuit.

Deep Neural Network models are integrated as parts of many real-world software applications. Due to the huge model size and complex computation, distributed deep learning (DDL) framework aims to provide a high-quality cluster scheduler to manage DDL training jobs from both resource allocation and job scheduling. However, existing schedulers either allocate a fixed amount of resources, or lack the control over task placement, which lead less efficient training. In this paper, we propose DeepSys, a GPU cluster scheduler tailored for DDL jobs. For single model, DeepSys builds a speed model to predict accurate training speed, and a memory model for high-quality resource utilization. For job scheduling, DeepSys considers resource allocation and task placement to provide efficient job scheduling in cluster. Experiments implemented on Kubernetes in two clusters show the advantage to the compared methods by 20% - 25% and 10% - 15% on average job completion time and makespan, respectively.