Professor Yunxin Liu from the Academy of Intelligent and Connected Systems (AIR) at Tsinghua University has had four papers accepted to MobiCom 2021, making him the author with the highest number of papers accepted this year. MobiCom, the International Conference on Mobile Computing and Networking, is a top-tier academic conference in the field of mobile computing and networking, and it is classified as an A-class conference by the China Computer Federation (CCF). The 27th edition of MobiCom, originally scheduled for 2021, was postponed due to the pandemic and held from March 28 to April 1, 2022, in a hybrid format. This conference accepted a total of 60 papers, with China and the United States leading in the number of accepted papers, each contributing 22. Among the contributing institutions, Tsinghua University and Microsoft Corporation were tied for the most papers accepted, with six each. The four papers accepted to MobiCom 2021 by Professor Liu focus on the field of intelligent edge computing, a research area he has dedicated himself to in recent years. These works were collaborations with institutions such as Microsoft Research, Nanjing University, University of Science and Technology of China, Xi'an Jiaotong University, Shanghai Jiaotong University, and Rutgers University in the United States. 1. **AsyMo: Scalable and Efficient Deep-Learning Inference on Asymmetric Mobile CPUs** - **Authors:** Man Nie, Shaohua Ding, Ting Cao, Yunxin Liu, Fenyuan Xu - **Summary:** This paper addresses the inefficiencies in deep learning inference on mobile CPUs, particularly those with asymmetric architectures (big and little cores). The authors identify that existing inference implementations fail to utilize the hardware resources effectively, leading to low core utilization and wasted resources. They propose AsyMo, a solution that optimizes task allocation and scheduling across big and little cores to maximize CPU utilization and reduce latency. AsyMo also includes an optimal energy consumption frequency setting strategy to minimize power usage during inference, thereby improving both performance and energy efficiency. 2. **Elf: Accelerate High-resolution Mobile Deep Vision with Content-aware Parallel Offloading** - **Authors:** Wuyang Zhang, Zhezhi He, Luyang Liu, Zhenhua Jia, Yunxin Liu, Marco Gruteser, Dipankar Raychaudhuri, Yan Yong Zhang - **Summary:** The Elf system aims to accelerate high-resolution deep vision tasks on mobile devices by leveraging parallel offloading to multiple edge cloud servers. The authors highlight that uniform image or video segmentation can degrade model accuracy and waste resources. Elf uses a lightweight attention LSTM network to predict and track regions of interest in video frames, ensuring that objects are not fragmented. The system then segments the video content based on computational complexity and network dynamics, distributing tasks efficiently among edge cloud servers to enhance performance without compromising accuracy. 3. **PECAM: Privacy-Enhanced Video Streaming and Analytics via Securely-Reversible Transformation** - **Authors:** Hao Wu, Xuejin Tian, Minghao Li, Yunxin Liu, Genesh Ananthanarayanan, Fenyuan Xu, Sheng Zhong - **Summary:** PECAM is a privacy-enhancing technology designed for video streaming and analytics (VSA) systems. It transforms video content in real-time to protect visual privacy, making it suitable for applications like smart cities, home security, and elderly care. The system uses a secure and reversible deep learning-based model that can hide sensitive details, such as license plates, while maintaining the ability to recognize objects. When authorized, the protected video can be restored to its original form for forensic purposes. PECAM also optimizes bandwidth usage, achieving 1.8 times the efficiency of H.264, and significantly outperforms popular models like CycleGAN and YoloV3 in terms of computational speed. 4. **Remix: Flexible High-resolution Object Detection on Edge Devices with Tunable Latency** - **Authors:** Shiqi Jiang, Zhiqi Lin, Yuanchun Li, Yuanchao Shu, Yunxin Liu - **Summary:** Remix is a flexible and tunable framework for high-resolution object detection on edge devices. The paper addresses the challenge of running computationally intensive deep neural networks (DNNs) on devices with limited resources, such as smart cameras. By analyzing and utilizing the diversity of existing models, Remix dynamically allocates computational resources to achieve optimal performance within specified latency constraints. The framework can speed up inference by up to 8.1 times compared to state-of-the-art object detection models while maintaining or even improving detection accuracy. These contributions by Professor Liu and his collaborators highlight significant advancements in the field of intelligent edge computing, addressing issues of efficiency, privacy, and performance in mobile and edge computing environments. The acceptance of these papers by MobiCom 2021 underscores the high quality and impact of the research being conducted at Tsinghua University's AIR.