Recent advancements in on-device training for deep neural networks have underscored the critical need for efficient activation compression to overcome the memory constraints of mobile and edge devices. As activations dominate memory usage during training and are essential for gradient computation, compressing them without compromising accuracy remains a key research challenge. While existing methods for dynamic activation quantization promise theoretical memory savings, their practical deployment is impeded by system-level challenges such as computational overhead and memory fragmentation.

Minimizing bandwidth consumption while maintaining satisfactory visual quality becomes the holy grail of volumetric content delivery. However, due to the huge amount of 3D data to stream, the stringent latency requirement, and the high computational workload, achieving this ambitious goal could be challenging for mobile mixed reality headsets, which can naturally enable viewers' motion with six degrees of freedom but have limited computing power. Motivated by our critical observations from a user study of eye movements with 50+ participants, in this paper, we present Theia, a first-of-its-kind gaze-driven and perception-aware volumetric content delivery system that effectively incorporates the following innovations into a holistic system: (1) real-time creation of foveated volumetric content to reduce network data usage; (2) efficient augmentation of foveal content to boost user experience; and (3) adaptive omission of peripheral content for further bandwidth savings based on eye movements. We implement a prototype of Theia using Microsoft HoloLens 2 headsets and extensively evaluate its performance. Our results reveal that compared to the state-of-the-art, Theia can drastically reduce bandwidth consumption by up to 67.0% and enhance visual quality by up to 92.5%.

Achieving the optimal balance of minimizing bandwidth consumption and end-to-end latency while preserving a satisfactory level of visual quality becomes the ultimate goal of live, interactive holographic communication, a fundamental building block of immersive telepresence envisioned for 6G. Nevertheless, achieving this ambitious goal poses significant challenges for mobile devices with limited computing power, considering the substantial amount of 3D data to stream, the demanding latency requirements, and the high computation workload involved. Instead of distributing immersive content bit by bit, in this position paper, we propose to deliver semantic information extracted from telepresence participants to drastically reduce Internet bandwidth usage for task-oriented applications such as remote collaboration. We contribute a taxonomy by categorizing related semantics into three different types (i.e., keypoints, 2D images, and text), pinpoint the open research challenges associated with developing a practical system for each category in our comprehensive research agenda, and delve into the potential solutions for overcoming these challenges. The preliminary results from our proof-of-concept implementation that harnesses keypoint-based semantics (partially) validate the feasibility of our research agenda.

By combining various emerging technologies, mobile extended reality (XR) blends the real world with virtual content to create a spectrum of immersive experiences. Although Web-based XR can offer attractive features such as better accessibility, cross-platform compatibility, and instant updates, its performance may not be on par with its standalone counterpart. As a low-level bytecode, WebAssembly has the potential to drastically accelerate Web-based XR by enabling near-native execution speed. However, little has been known about how well Web-based XR performs with WebAssembly acceleration. To bridge this crucial gap, we conduct a first-of-its-kind systematic and empirical study to analyze the performance of Web-based XR expedited by WebAssembly on four diverse platforms with five different browsers. Our measurement results reveal that although WebAssemlby can accelerate different XR tasks in various contexts, there remains a substantial performance disparity between Web-based and standalone XR. We hope our findings can foster the realization of an immersive Web that is accessible to a wider audience with various emerging technologies.