XLXuandong Liu

Xuandong Liu

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Xuandong Liu has a total of 20 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Optimization Design of Shielding Cover for Valve-Side Bushings Considering Interface Charge and Electric Field Distributions

    by Minjie Li, Gaoyi Shang, WanHao SHI, Hao Sun, Xueyang Bai, Hao Tang, Yi Zhang, Xuandong Liu

    Frequent gas-solid interface flashover accidents on valve-side bushings seriously threaten the safe operation of ultra-high voltage direct current (UHVDC) systems. However, optimizing their shielding cover structures still lacks systematic theoretical support. This paper establishes a electric-thermal-ion transport multi-physics coupled model of ±800 kV valve-side bushing. The influence mechanism of the shield cover structure (width, gap distance, and chamfer radius) on charge accumulation and electric field distortion at the core end is studied. Results show that the shield width is the dominant parameter affecting the charge and electric field peaks at the core end. Broadening width from 50 mm to 80 mm suppresses them by over 40% and the peak electric field shift toward the long surface. However, once the electric field peak fully appears on the long surface, further widening offers limited benefit. Meanwhile, gap distance and chamfer radius act locally, suppressing charge density and electric field at gap surface while rebounding the peaks at corner arc. Based on these variation law, a dimensionless design criterion incorporating a corner arc concentration factor (Kc) and a gap non-uniformity factor (Kgap) are proposed, which translate complex electric variation trends into empirical formulas. Ultimately, the optimal shield dimensions for a ±800 kV bushing are identified as w = 80 mm, d = 5–8 mm, and r = 5–30 mm.

  • AIRP: Accelerating Multi-Tenant Distributed Learning With In-Network Resource Pooling

    by Huifeng Xing, Hao Wang, Yang Chen, Yinfan Hu, Zixuan Chen, Xuandong Liu, Zijian Li, Wanxin Shi, Sen Liu, Yang Xu

    The increasing popularity of large models and datasets has highlighted the significance of distributed training networks. As gradient synchronization generates substantial traffic, in-network aggregation (INA) has emerged as a solution to offload aggregation onto the switch, alleviating network congestion and accelerating distributed training. However, the limited memory capacity of the INA switch becomes a potential bottleneck as computation shifts into the network, especially in multi-tenant scenarios. To address this bottleneck and enhance network throughput, we propose the Aggregation with Innetwork Resource Pooling (AIRP) framework. Unlike existing approaches that optimize individual switches in a localized manner, AIRP takes a holistic view and efficiently pools switch memory resources across the entire network, allocating them to multiple tenants. Evaluation using the ns-3 simulator and P4 testbed demonstrates that AIRP can accelerate the training of various models, including computer vision and language models. The experimental results show that AIRP outperforms existing INA approaches by up to 7 times in terms of network throughput in multi-tenant scenarios, while also achieving great flexibility and efficiency in deployment.

  • Open Access

    OSP: Boosting Distributed Model Training with 2-stage Synchronization

    by Zixuan Chen, Lei Shi, Xuandong Liu, Jiahui Li, Sen Liu, Yang Xu

    Distributed deep learning (DDL) is a promising research area, which aims to increase the efficiency of training deep learning tasks with large size of datasets and models. As the computation capability of DDL nodes continues to increase, the network connection between nodes is becoming a major bottleneck. Various methods of gradient compression and improved model synchronization have been proposed to address this bottleneck in Parameter-Server-based DDL. However, these two types of methods can result in accuracy loss due to discarded gradients and have limited enhancement on the throughput of model synchronization, respectively. To address these challenges, we propose a new model synchronization method named Overlapped Synchronization Parallel (OSP), which achieves efficient communication with a 2-stage synchronization approach and uses Local-Gradient-based Parameter correction (LGP) to avoid accuracy loss caused by stale parameters. The prototype of OSP has been implemented using PyTorch and evaluated on commonly used deep learning models and datasets with a 9-node testbed. Evaluation results show that OSP can achieve up to 50% improvement in throughput without accuracy loss compared to popular synchronization models.


  • Open Access

    Augmenting Distributed AI Training with Loss-tolerant Transmission

    by Zixuan Chen, Lei Shi, Yongbo Gao, Xuandong Liu, Xin Ai, Sen Liu, Yang Xu