Deploying deep neural networks (DNNs) in dynamic edge environments is challenging, as scheduling decisions must consider both the topological complexity of the DNN model and the variability of dynamic edge resources. Existing methods often fail to balance this trade-off, resulting in suboptimal performance. In this study, we introduce Heterogeneous Graph-Attention-Network-Based Collaborative Inference (HANCI), a framework designed to minimize end-to-end inference latency using an intelligent, context-aware scheduling approach. This framework employs a decoupled feature extraction architecture based on a heterogeneous graph attention network. It independently extracts high-quality features from static DNN computational graphs and dynamic edge environments. This decoupling allows HANCI to generate comprehensive, interference-free state embeddings. These embeddings support a hybrid decision-making strategy that combines a learned partitioning policy with a heuristic-guided placement mechanism, making the complex scheduling problem tractable. Experimental evaluations demonstrate that HANCI considerably outperforms baseline methods. It reduces inference latency by 21.6% for lightweight models and by up to 78.1% for complex architectures. This study provides a robust and adaptive solution to the challenge of collaborative DNN inference in edge environments.

Built on the WeChat social platform, WeChat Mini-Programs are widely used by more than 400 million users every day. Consequently, the reliability of Mini-Programs is particularly crucial. However, WeChat Mini-Programs suffer from various bugs related to execution environment, lifecycle management, asynchronous mechanism, etc. These bugs have seriously affected users' experience and caused serious impacts.

Node.js has become a widely-used event-driven architecture for server-side and desktop applications. Node.js provides an effective asynchronous event-driven programming model, and supports asynchronous tasks and multi-priority event queues. Unexpected races among events and asynchronous tasks can cause severe consequences. Existing race detection approaches in Node.js applications mainly adopt random fuzzing technique, and can miss races due to large schedule space.In this paper, we propose a dynamic race detection approach NRace for Node.js applications. In NRace, we build precise happens-before relations among events and asynchronous tasks in Node.js applications, which also take multi-priority event queues into consideration. We further develop a predictive race detection technique based on these relations. We evaluate NRace on 10 realworld Node.js applications. The experimental result shows that NRace can precisely detect 6 races, and 5 of them have been confirmed by developers.

JavaScript is one of the most popular programming languages. WeChat Mini-Program is a large ecosystem of JavaScript applications that runs on the WeChat platform. Millions of Mini-Programs are accessed by WeChat users every week. Consequently, the performance and robustness of Mini-Programs are particularly important. Unfortunately, many Mini-Programs suffer from various defects and performance problems. Dynamic analysis is a useful technique to pinpoint application defects. However, due to the dynamic features of the JavaScript language and the complexity of the runtime environment, dynamic analysis techniques were rarely used to improve the quality of JavaScript applications running on industrial platforms such as WeChat Mini-Program previously. In this work, we report our experience of extending Jalangi, a dynamic analysis framework for JavaScript applications developed by academia, and applying the extended version, named WeJalangi, to diagnose defects in WeChat Mini-Programs. WeJalangi is compatible with existing dynamic analysis tools such as DLint, Smemory, and JITProf. We implemented a null pointer checker on WeJalangi and tested the tool's usability on 152 open-source Mini-Programs. We also conducted a case study in Tencent by applying WeJalangi on six popular commercial Mini-Programs. In the case study, WeJalangi accurately located six null pointer issues and three of them haven't been discovered previously. All of the reported defects have been confirmed by developers and testers.