Gait recognition is emerging as a promising technology and an innovative field within computer vision, with a wide range of applications in remote human identification. However, existing methods typically rely on complex architectures to directly extract features from images and apply pooling operations to obtain sequence-level representations. Such designs often lead to overfitting on static noise (e.g., clothing), while failing to effectively capture dynamic motion regions, such as the arms and legs. This bottleneck is particularly challenging in the presence of intra-class variation, where gait features of the same individual under different environmental conditions are significantly distant in the feature space. To address the above challenges, we present a Language-guided and Motion-aware gait recognition framework, named LMGait. To the best of our knowledge, LMGait is the first method to introduce natural language descriptions as explicit semantic priors into the gait recognition task. In particular, we utilize designed gait-related language cues to capture key motion features in gait sequences. To improve cross-modal alignment, we propose the Motion Awareness Module (MAM), which refines the language features by adaptively adjusting various levels of semantic information to ensure better alignment with the visual representations. Furthermore, we introduce the Motion Temporal Capture Module (MTCM) to enhance the discriminative capability of gait features and improve the model’s motion tracking ability. We conducted extensive experiments across multiple datasets, and the results demonstrate the significant advantages of our proposed network. Specifically, our model achieved accuracies of 88.5%, 97.1%, and 97.5% on the CCPG, SUSTech1K, and CASIAB* datasets, respectively, achieving state-of-the-art performance.

Score Distillation Sampling (SDS) enables 3D asset generation by distilling priors from pretrained 2D text-to-image diffusion models, but vanilla SDS suffers from over-saturation and over-smoothing. To mitigate this issue, recent variants have incorporated negative prompts. However, these methods face a critical trade-off: limited texture optimization, or significant texture gains with shape distortion. In this work, we first conduct a systematic analysis and reveal that this trade-off is fundamentally governed by the utilization of the negative prompts, where Target Negative Prompts (TNP) that embed target information in the negative prompts dramatically enhancing texture realism and fidelity but inducing shape distortions. Informed by this key insight, we introduce the Target-Balanced Score Distillation (TBSD). It formulates generation as a multi-objective optimization problem and introduces an adaptive strategy that effectively resolves the aforementioned trade-off. Extensive experiments demonstrate that TBSD significantly outperforms existing state-of-the-art methods, yielding 3D assets with high-fidelity textures and geometrically accurate shape.

Masked Image Modeling (MIM) with Vector Quantization (VQ) has achieved great success in both self-supervised pre-training and image generation. However, most existing methods struggle to address the trade-off in shared latent space for generation quality vs. representation learning and efficiency. To push the limits of this paradigm, we propose MergeVQ, which incorporates token merging techniques into VQ-based generative models to bridge the gap between image generation and visual representation learning in a unified architecture. During pre-training, MergeVQ decouples top-k semantics from latent space with the token merge module after self-attention blocks in the encoder for subsequent Look-up Free Quantization (LFQ) and global alignment and recovers their fine-grained details through cross-attention in the decoder for reconstruction. As for second-stage generation, we introduce MergeAR, which performs KV Cache compression for efficient raster-order prediction. Extensive experiments on ImageNet verify that MergeVQ as an AR generative model achieves competitive performance in both visual representation learning and image generation tasks while maintaining favorable token efficiency and inference speed. Code and model will be available at https://apexgen-x.github.io/MergeVQ.

Achieving higher spectrum efficiency has been a critical performance target for the forthcoming 6G era. Non-orthogonal multiple access (NOMA) is regarded as an effective technique to improve the spectrum efficiency a nd t he cluster-free NOMA is a novel generalized downlink NOMA transmission framework. However, there still lacks an efficient joint beamforming and SIC optimization algorithm for cluster-free NOMA. In this paper, we propose a novel channel correlation based two-loop greedy (CC-TLG) algorithm to obtain a near-optimal solution with dramatically reduced complexity. CC-TLG maximizes the WSR of users based on the user channel correlation coefficients, where the outer loop add users one by one based on the WSR in a greedy manner, and the inner loop generates a near-optimal SIC operation based on both the channel correlation coefficients and WSR in a greedy way. Finally, simulations demonstrates the effectiveness of the proposed CC-TLG algorithm.

In the forthcoming 6G era, extend reality (XR) has been regarded as an emerging application for ultra-reliable and low latency communications (URLLC) with new traffic characteristics and more stringent requirements. In addition to the quasi-periodical traffic in XR, burst traffic with both large frame size and random arrivals in some real world low latency communication scenarios has become the leading cause of network congestion or even collapse, and there still lacks an efficient algorithm for the resource scheduling problem under burst traffic with hard latency constraints. We propose a policy reuse reinforcement learning framework for resource scheduling with hard latency constraints (PRRL-RSHLC), which maximizes the hard-latency constrained effective throughout (HLC-ET) of users. The proposed algorithm reuses polices from both old policies learned under other similar environments and domain-knowledge-based (DK) policies constructed using expert knowledge to improve the performance. Simulations show that PRRL-RSHLC can achieve superior performance with faster convergence speed compared to baseline aIgorithms.

A novel hard-latency guaranteed cluster-free multiple-input multiple-output non-orthogonal multiple access (MIMO-NOMA) framework is proposed to deal with burst traffics that commonly occur in real-world scenarios. The hard-latency constrained effective throughput (HLC-ET) maximization problem is formulated, which jointly optimizes the beamforming and cluster-free success interference cancellation (SIC) operations. To address the resultant problem, a two-stage reinforcement learning (RL)-based algorithm is developed to capture system uncertainty, where the large-dimension optimization is decoupled into two stages to reduce the action space and fasten convergence of RL. In the long-term stage, we aim to maximize the HLC-ET, and a hybrid RL algorithm with policy reuse is adoped to control the priority weights to construct the weighted sum rate (WSR) function of users. In the short-term stage, a branch-and-bound (BB) based algorithm is further developed to obtain the optimal solution of the WSR maximization problem. The BB-based algorithm is proved to guarantee the convergence to an ϵ-optimal solution of the WSR maximization problem within a finite number of steps. To accelerate computation in the short-term stage, a channel correlation based two-loop greedy (CC-TLG) algorithm is proposed to significantly reduce the complexity with almost no performance loss compared to the BB-based algorithm. Finally, simulations demonstrate the advantages of the proposed two-stage RL based joint beamforming and SIC optimization (TSRL-JBSO) algorithm over conventional RL-based and non-RL based algorithms.

Multi-user MIMO (MU-MIMO) enables a base station (BS) to transmit data streams to multiple users simultaneously on the same resource block (RB). In 5G specifications, a resource scheduling algorithm needs to consider both resource allocation in the spatial domain (i.e., MU-MIMO user scheduling on each RB) and frequency domain (i.e., RB allocation to each user). In addition, the scheduler must meet the real-time requirement. This paper presents a random sampling based joint spatial-frequency domain message passing (RS-JSFD-MP) algorithm, which is a novel radio resource scheduler that can meet the real-time requirement under the consideration of the finite buffer traffic scenario. The key idea of RS-JSFD-MP is to transform the scheduling problem into a graph model and design the corresponding low complexity message passing algorithm based on the random sampling approach. Experimental results show that RS-JSFD-MP can achieve better scheduling performance than existing greedy baseline algorithm, and can also obtain lower complexity by choosing an appropriate number of random samples. Moreover, RS-JSFD-MP facilitates parallel and distributed implementation, which helps to accelerate computation time to meet the real-time requirement on the resource scheduling algorithm.

In this study, secure communications in a two‐way relaying wiretap system with a silent eavesdropper are studied. Under the assumption that the channel state information relating to the eavesdropper is unknown at legitimate nodes, the precoding vectors at the two source nodes are jointly designed to decrease the wiretap capacity at the eavesdropper. The derived relationship between the two precoding vectors is partially similar in nature to the linear minimum mean squared error (MMSE) equaliser. An iterative process is proposed to jointly achieve two precoding vectors, and its convergence is proved. Simulation results are presented to demonstrate that the proposed MMSE‐based precoding algorithm can be used for source nodes with any number of antennas, and outperforms its conventional zero‐forcing‐ and matched‐filter‐based counterparts. Signal leakage from the relay node to the eavesdropper is nearly completely suppressed by the proposed precoding design.