Weijie Shi, Jipeng Zhang, Yaguang Wu, Jingzhi Fang, Shibo Zhang, Yao Zhao, Hao Chen, Ruiyuan Zhang, Yue Cui, Jia Zhu, Sirui Han, Jiajie Xu, Xiaofang Zhou. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Recently, text-to-image models based on diffusion have achieved remarkable success in generating high-quality images. However, the challenge of personalized, controllable generation of instances within these images remains an area in need of further development. In this paper, we present LocRef-Diffusion, a novel, tuning-free model capable of personalized customization of multiple instances’ appearance and position within an image. To enhance the precision of instance placement, we introduce a Layout-net, which controls instance generation locations by leveraging both explicit instance layout information and an instance region cross-attention module. To improve the appearance fidelity to reference images, we employ an appearance-net that extracts instance appearance features and integrates them into the diffusion model through cross-attention mechanisms. We conducted extensive experiments on the COCO and OpenImages datasets, and the results demonstrate that our proposed method achieves state-of-the-art performance in layout and appearance guided generation.

Diverse decoding of large language models is crucial for applications requiring multiple semantically distinct responses, yet existing methods primarily achieve lexical rather than semantic diversity. This limitation significantly constrains Best-of-N strategies, group-based reinforcement learning, and data synthesis. While temperature sampling and diverse beam search modify token distributions or apply n-gram penalties, they fail to ensure meaningful semantic differentiation. We introduce Semantic-guided Diverse Decoding (SemDiD), operating directly in embedding space that balances quality with diversity through three complementary mechanisms: orthogonal directional guidance, dynamic inter-group repulsion, and position-debiased probability assessment. SemDiD harmonizes these competing objectives using adaptive gain functions and constraint optimization, ensuring both quality thresholds and maximal semantic differentiation. Experiments show SemDiD consistently outperforms existing methods, improving Best-of-N coverage by 1.4-5.2% across diverse tasks and accelerating RLHF training convergence by 15% while increasing accuracy by up to 2.1%.

As traditional wireless cellular networks are facing a rapid growth of traffic demand, future networks may consist of a large number of small cells which are densely deployed. However, networks in dense environment is exposed to strong inter-cell interference problem which is critical to networks capacity. To solve the problem, we use the Graph Coloring Algorithm (GCA) to divide the small cells into different clusters to mitigate the serious inter-cell interference between the dense small cells, and accordingly reducing the transmission power of small cell BSs in some subframes based on the proposed utility function. The existence of the optimal solutions to the problem is discussed, then the Differential Evolution (DE) is applied to search the optimal transmission power which maximizes the proposed utility function. Intensive simulations show that the proposed scheme can significantly mitigate the co-tier interference and maximize networks capacity.

Dense deployment of small cell in heterogeneous networks (HetNet) is considered as a promising way to cope with the exponential growth in mobile traffic demand. In such scenario, coverage optimization is a challenge due to the unplanned deployment and plug-and-play feature of small cell. In this paper, the coverage performance of the network is analyzed under different deployment density of small cells. Then the coverage problem is modeled as a cost function which includes the inappropriate coverage ratio of the network. After that, the ant colony algorithm (ACA) is applied to find the optimal pilot Tx power of each small cell through the minimization of the cost function. Finally, compared to the fixed coverage scheme, the proposed algorithm can not only reduce the ratio of coverage hole but also the proportion of coverage overlap as well.

Improving network Energy Efficiency (EE) in wireless communications has attracted increasing attention recently. The purpose of our research is to optimize the network power consumption and preserve the throughput of macrocells User Equipments (UEs) at the same time. We solve the problem by controlling the transmission power of macro base stations (BSs) in some subframes based on the proposed utility function, and accordingly scheduling picocells Cell Range Expansion (CRE) UEs in these subframes. The existence of the optimal solutions to the problem is discussed, then the Differential Evolution (DE) is applied to search the optimal transmission power which maximizes the proposed utility function. System level simulation results show that the proposed algorithm can not only optimize the network power consumption but also improve the throughput of the network, which will lead to the network EE improvement.

In Heterogeneous Network (HetNet), reasonable resource allocation has attracted extensive attention. To allocate the resource more effectively, we propose a novel algorithm from the perspective of both time domain and power domain. On the one hand, we propose an adaptive Almost Blank Subframe (ABS) configuration scheme to dynamically match the network resources with the real-time load. On the other hand, we propose a utility function of macrocells' power control and a corresponding scheduling scheme to make a tradeoff between the two-tier macro-pico networks and protect the victim users as well. The existence of the optimal solution to the problem is proved, and Differential Evolution (DE) is applied to find the optimal solution. System level simulation results show that the proposed algorithm can not only enhance the load balance between macrocell and picocell, but also provide a great improvement on the performance of edge users with little overall throughput cost.