International audience

Combinatorial multi-armed bandit (CMAB) model is designed to maximize cumulative rewards in the presence of uncertainty by selecting a subset of arms in each round. This paper is inspired by two critical applications in wireless networks, where it’s not only essential to maximize cumulative rewards but also to guarantee fairness among arms (i.e., the minimum average reward required by each arm) and ensure reward regularity (i.e., how often each arm receives the reward). In this paper, we propose a parameterized regular and fair learning algorithm to achieve these three objectives. In particular, the proposed algorithm linearly combines virtual queue-lengths (tracking the fairness violations), Time-Since-Last-Reward (TSLR) metrics, and Upper Confidence Bound (UCB) estimates in its weight measure. Here, TSLR is inspired by age-of-information and measures the elapsed number of rounds since an arm last received a reward, capturing the reward regularity performance, and UCB estimates are utilized to balance the tradeoff between exploration and exploitation in online learning. By uncovering a key relationship between the dynamics of virtual queue-lengths and TSLR metrics and utilizing several non-trivial Lyapunov functions, we analytically characterize zero cumulative fairness violation, reward regularity, and cumulative regret performance under our proposed algorithm. These theoretical outcomes are verified by simulations based on two real-world datasets.

A comprehensive understanding of 3D scenes is essential for autonomous vehicles (AVs), and among various perception tasks, occupancy estimation plays a central role by providing a general representation of drivable and occupied space. However, most existing occupancy estimation methods rely on LiDAR or cameras, which perform poorly in degraded environments such as smoke, rain, snow, and fog. In this paper, we propose 4D-ROLLS, the first weakly supervised occupancy estimation method for 4D radar using the LiDAR point cloud as the supervisory signal. Specifically, we introduce a method for generating pseudo-LiDAR labels, including occupancy queries and LiDAR height maps, as multi-stage supervision to train the 4D radar occupancy estimation model. Then the model is aligned with the occupancy map produced by LiDAR, fine-tuning its accuracy in occupancy estimation. Extensive comparative experiments validate the exceptional performance of 4D-ROLLS. Its robustness in degraded environments and effectiveness in cross-dataset training are qualitatively demonstrated. The model is also seamlessly transferred to downstream tasks BEV segmentation and point cloud occupancy prediction, highlighting its potential for broader applications. The lightweight network enables 4D-ROLLS model to achieve fast inference speeds at about 30 Hz on a 4060 GPU. The code of 4D-ROLLS will be made available at https://github.com/CLASS-Lab/4D-ROLLS.

Immersive technologies, such as virtual and augmented reality, demand high framerate, low latency, and precise synchronization between real and virtual environments. To meet these requirements, an edge server typically needs to perform high-quality rendering, and must predict user head motion and transmit a portion of the rendered panoramic scene that is large enough to cover the user's viewport, yet small enough to satisfy bandwidth constraints. Each portion yields two feedback signals: prediction feedback, indicating whether the selected portion covers the actual viewport, and transmission feedback, indicating whether all data packets are successfully delivered. While prior work models this setting as a multi-armed bandit with two-level bandit feedback, it overlooks that prediction feedback can be retrospectively computed for all possible portions, thus providing full-information feedback. In this work, we introduce a new two-level feedback model that combines full-information feedback with bandit feedback, and we formulate the portion selection problem as an online learning task under this hybrid setting. We derive an instance-dependent regret lower bound for this new hybrid feedback setting, and we propose AdaPort, a hybrid learning algorithm that leverages both the full-information feedback and bandit feedback to improve learning efficiency. We then show that the instance-dependent regret upper bound for AdaPort matches the lower bound asymptotically, proving its asymptotic optimality. Simulations using synthetic data and real-world traces demonstrate that AdaPort consistently outperforms state-of-the-art baselines, validating the benefits of exploiting the hybrid feedback structure.

Wireless interactive panoramic scene delivery imposes unique challenges compared to its wired or non-interactive counterparts. The wireless channel is throughput-constrained, which limits the ability to deliver large and high-quality panoramic images. On the other hand, the interactivity imposes a real-time constraint on the system and limits the use of a playback buffer. Also, wireless inputs are not delivered instantaneously like wired interrupt-based inputs, so the system must predict the user's head pose and the portion of the scene visible to them, called the viewport. This reveals a tradeoff: delivering a portion too small may not cover the viewport if the prediction error is too large, while delivering a portion too large may result in a failed wireless transmission. Likewise, delivering the portion at too high a quality may result in a failed transmission. Despite these challenges, we would like to guarantee an immersive experience for the user by delivering a high-quality and visually consistent panoramic scene. To that end, we aim to maximize the user's quality of experience, which we define as the combination of (1) cumulative quality, (2) long-term consistency, which quantifies the overall variance in perceived quality, and (3) short-term consistency, which quantifies abrupt quality changes. We formulate this problem as a risk-averse multi-armed bandit problem with reward-dependent switching costs, and develop a novel block-based UCB algorithm with opportunistic switching. We derive its theoretical regret upper bound, which matches results in prior work, and corroborate this result in trace-based simulations using a panoramic video streaming data trace.

The integration of Large Language Models (LLMs) like GPT-4 with Extended Reality (XR) technologies offers the potential to build truly immersive XR environments that interact with human users through natural language, e.g., generating and animating 3D scenes from audio inputs. However, the complexity of XR environments makes it difficult to accurately extract relevant contextual data and scene/object parameters from an overwhelming volume of XR artifacts. It leads to not only increased costs with pay-per-use models, but also elevated levels of generation errors. Moreover, existing approaches focusing on coding script generation are often prone to generation errors, resulting in flawed or invalid scripts, application crashes, and ultimately a degraded user experience. To overcome these challenges, we introduce LLMER, a novel framework that creates interactive XR worlds using JSON data generated by LLMs. Unlike prior approaches focusing on coding script generation, LLMER translates natural language inputs into JSON data, significantly reducing the likelihood of application crashes and processing latency. It employs a multi-stage strategy to supply only the essential contextual information adapted to the user's request and features multiple modules designed for various XR tasks. Our preliminary user study reveals the effectiveness of the proposed system, with over 80% reduction in consumed tokens and around 60% reduction in task completion time compared to state-of-the-art approaches. The analysis of users' feedback also illuminates a series of directions for further optimization.

Combinatorial multi-armed bandit refers to the model that aims to maximize cumulative rewards in the presence of uncertainty. Motivated by two important wireless network applications, in addition to maximizing cumulative rewards, it is important to ensure fairness among arms (i.e., the minimum average reward required by each arm) and reward regularity (i.e., how often each arm receives the reward). In this paper, we develop a parameterized regular and fair learning algorithm to achieve these three objectives. In particular, the proposed algorithm linearly combines virtual queue-lengths (tracking the fairness violations), Time-Since-Last-Reward (TSLR) metrics, and Upper Confidence Bound (UCB) estimates in its weight measure. Here, TSLR is similar to age-of-information and measures the elapsed number of rounds since the last time an arm received a reward, capturing the reward regularity performance, and UCB estimates are utilized to balance the tradeoff between exploration and exploitation in online learning. Through capturing a key relationship between virtual queue-lengths and TSLR metrics and utilizing several non-trivial Lyapunov functions, we analytically characterize zero cumulative fairness violation, reward regularity, and cumulative regret performance under our proposed algorithm. These findings are corroborated by our extensive simulations.

The Internet of Things (IoT) interconnects a vast number of physical devices with each other and enables remote monitoring and control of the physical system. This, together with the recent progress of virtual reality (VR) technology, provides an immersive experience for the user to perform remote monitoring and control as if she observes and controls the physical system in person. In this demo, we develop an immersive remote monitoring and control system consisting of 3D virtual monitoring and control panel and the physical water pump system. Users can remotely control and monitor the water level of the physical water pump system through the 3D virtual panel in real-time. We use Fourier transform to filter out the noise when the water level remains static and the exponential running average method to make the collected data more smooth when the water level dynamically changes. The experimental evaluations demonstrate that the difference between the physical and virtual water levels is small in both static and dynamic water levels.

Rate adaptation is a key mechanism in current IEEE 802.11 networks and next-generation cellular systems. Observing that the operating time scale of rate adaptation is usually much smaller than the user association and scheduling, we study a joint design of wireless user association and scheduling and rate adaptation with different time scales to maximize cumulative system throughput while guaranteeing desired fairness among users. We develop a maximum-weight type user association and scheduling algorithm that combines the virtual queues (tracking the scheduling debt for each user to ensure the desired fairness guarantee) and Upper Confidence Bound (UCB) estimates in its weight measure; each selected user then adopts the UCB algorithm to perform rate adaptation in a smaller time scale. We show that our proposed algorithm yields a cumulative regret growing with the square root of the time horizon up to a logarithmic factor, and achieves zero cumulative fairness violation after a certain number of time frames. We demonstrate the efficiency of the proposed algorithm via simulations using synthetic and realistic data traces.

The emergence of unmanned aerial vehicles (UAVs) extends the mobile edge computing (MEC) services in broader coverage to offer new flexible and low-latency computing services for user equipment (UE) in the era of 5G and beyond. One of the fundamental requirements in UAV-assisted mobile wireless systems is the low latency, which can be jointly optimized with service caching and task offloading. However, this is challenged by the communication overhead involved with service caching and constrained by limited energy capacity. In this work, we present a comprehensive optimization framework with the objective of minimizing the service latency while incorporating the unique features of UAVs. Specifically, to reduce the caching overhead, we make caching placement decision every T slots (specified by service providers), and adjust UAV trajectory, user equipment or UE-UAV association, and task offloading decisions at each time slot under the constraints of UAV’s energy and resource capacity. By leveraging Lyapunov optimization approach and dependent rounding technique, we design an alternating optimization-based algorithm, named TJSO, which iteratively optimizes caching and offloading decisions. Theoretical analysis proves that TJSO converges to the near-optimal solution in polynomial time. Extensive simulations further verify that our proposed solution can significantly reduce the service delay for UEs while maintaining low energy consumption when compared to the three state-of-the-art baselines.