With the increasing popularity of virtual reality (VR) applications, allowing interaction with non-immersed spectators has become an important supporting functionality. To this end, a variety of VR streaming systems focus on allowing remote spectating experiences. However, such systems lack mechanisms that allow spectators to assist or explore inside VR applications alongside the VR players. To address, we explore AccompliceVR (AVR), a novel streaming and remote collaboration system. Using built-in OpenVR functionality, AVR introduces avatar representations of spectators overlaid into the normal stereo views of unmodified VR applications. AVR also enhances the remote spectator experience by stitching frames of video from the VR player to reconstruct a pseudo-panoramic view. This allows them to move around and point at the environment with an avatar. We first evaluated that AVR is compatible with a selection of the most popular VR games. The system is then validated through a paired user study with 30 participants using a mixed-design pattern. Participants played two well-known VR games (as 'pilot') and were assisted by their partner in the pair (as 'copilot'). The results support that AVR shows significant improvement in communication between pilots and copilots and enhances the pilot's understanding of instructions from their copilot.

Deep Neural Networks (DNNs), despite their notable progress across information retrieval tasks, encounter the issues of shortcut learning and struggle with poor generalization due to their reliance on spurious correlations between features and labels. Current research mainly mitigates shortcut learning behavior using augmentation and distillation techniques, but these methods could be laborious and introduce unwarranted biases. To tackle these, in this paper, we propose COMI, a novel method to COrrect and MItigate shortcut learning behavior. Inspired by the ways students solve shortcuts in educational scenarios, we aim to reduce model's reliance on shortcuts and enhance its ability to extract underlying information integrated with standard Empirical Risk Minimization (ERM). Specifically, we first design Correct Habit (CoHa) strategy to retrieve the top m challenging samples for priority training, which encourages model to rely less on shortcuts in the early training. Then, to extract more meaningful underlying information, the information derived from ERM is separated into task-relevant and task-irrelevant information, the former serves as the primary basis for model predictions, while the latter is considered non-essential. However, within task-relevant information, certain potential shortcuts contribute to overconfident predictions. To mitigate this, we design Deep Mitigation (DeMi) network with shortcut margin loss to adaptively control the feature weights of shortcuts and eliminate their influence. Besides, to counteract unknown shortcut tokens issue in NLP, we adopt locally interpretable module-LIME to help recognize shortcut tokens. Finally, extensive experiments conducted on NLP and CV tasks demonstrate the effectiveness of COMI, which can perform well on both IID and OOD samples.