This page has only limited features, please log in for full access.
Xiaqing Li received his Ph.D degree in computer science from Tsinghua University in 2020. He is now a lecturer in Dept. of Computer Science and Technology at Beijing University of Technology. His research lies in the broad field of high performance computing, Networks-on-Chip, GPGPU and deep learning acceleration. His papers are published in major top journals and important conferences of the computer system, such as IEEE Transactions on Parallel and Distributed Systems (TPDS) and International Conference on Parallel Processing (ICPP). Now, he is also a reviewer for TPDS.
Routing algorithms is a key factor that determines the performance of NoC (Networks-on-Chip) systems. Regional congestion awareness routing algorithms have shown great potential in improving the performance of NoC. However, it incurs a significant queuing latency when practitioners use existing regional congestion awareness routing algorithms to make routing decisions, thus degrading the performance of NoC. In this paper, we propose an efficient area partition-based congestion-aware routing algorithm, ParRouting, which aims at increasing the throughput and reducing the latency for NoC systems. First, ParRouting partitions the network into two areas (i.e., edge area and central area.) based on node priorities. Then, for the edge area, ParRouting selects the output node based on different priorities for higher throughput; for the central area, ParRouting selects the node in the low congestion direction as the output node for lower queuing latency. Our experimental results indicate that ParRouting achieves a 53.4% reduction in packet average latency over SPLASH -2 ocean application and improves the saturated throughput by up to 38.81% over a synthetic traffic pattern for an NoC system, compared to existing routing algorithms.
Juan Fang; Di Zhang; Xiaqing Li. ParRouting: An Efficient Area Partition-Based Congestion-Aware Routing Algorithm for NoCs. Micromachines 2020, 11, 1034 .
AMA StyleJuan Fang, Di Zhang, Xiaqing Li. ParRouting: An Efficient Area Partition-Based Congestion-Aware Routing Algorithm for NoCs. Micromachines. 2020; 11 (12):1034.
Chicago/Turabian StyleJuan Fang; Di Zhang; Xiaqing Li. 2020. "ParRouting: An Efficient Area Partition-Based Congestion-Aware Routing Algorithm for NoCs." Micromachines 11, no. 12: 1034.
Deploying a practical ConvNet system requires not only high inference accuracy, but also small inference memory and fast training speed. However, existing approaches of hyper-parameter tuning only focus on high accuracy. Although achieving a high accuracy, tuning poorly can significantly increase the performance burden, and thus degrade the overall performance of a ConvNet system. In this paper, we propose SmartTuning, an approach to identifying the hyper-parameters of a ConvNet system for high training speed and low working memory, with the restriction of high inference accuracy. The core idea of SmartTuning is to build a new performance model for a ConvNet system, and then to integrate Bayesian Optimization to learn the relationship between the objective function and the hyperparameters in a CNN model. In this way, SmartTuning can balance the inference accuracy, inference memory usage and training speed during the tuning process, and thus maximizes the overall performance of a ConvNet system. Our experiments show that SmartTuning can stably identify the hyper-parameter sets that offer the same high-level accuracy while much less inference memory usage (i.e., ranges from 18 times to 24 times over MNIST and from 4 times to 9 times over CIFAR-10.) and faster training speed (i.e., ranges from 7 times to 11 times over MNIST and from 2 times to 3 times over CIFAR-10.), compared with existing tuning approaches.
Xiaqing Li; Guangyan Zhang; Weimin Zheng. SmartTuning: Selecting HyperParameters of a ConvNet System for Fast Training and Small Working Memory. IEEE Transactions on Parallel and Distributed Systems 2020, PP, 1 -1.
AMA StyleXiaqing Li, Guangyan Zhang, Weimin Zheng. SmartTuning: Selecting HyperParameters of a ConvNet System for Fast Training and Small Working Memory. IEEE Transactions on Parallel and Distributed Systems. 2020; PP (99):1-1.
Chicago/Turabian StyleXiaqing Li; Guangyan Zhang; Weimin Zheng. 2020. "SmartTuning: Selecting HyperParameters of a ConvNet System for Fast Training and Small Working Memory." IEEE Transactions on Parallel and Distributed Systems PP, no. 99: 1-1.