ID 原文 译文
25295 而在路由方案的优化过程中，准确预估给定路由方案下的网络性能是其关键。 In the process of optimizing a routing scheme, it is the key to accurately predict the network performance under a given routing scheme．
25296 本文基于图神经网络建模网络中物理链路与路由方案路径的关系，在给定的路由方案与网络流量下对网络中的各项端到端性能指标(如延迟、抖动)进行准确预估，以辅助优化路由方案。 This paper uses graph neural networks to model the relationship between physical links and routing scheme paths, so that the model can predict various end-to-end performance indicators (such as delay and jitter) in the network under a given routing scheme and network traffic．
25297 本文基于 OMNeT + + 来生成数据并进行实验，实验结果表明本文提出的模型能够针对延迟抖动等端到端性能指标进行准确预估，预估平均相对误差不超过 4. 1% 。 This paper uses OMNeT+ + to generate datasets. The experimental results show that the model proposed in this paper can accurately predict end-to-end performance indicators such as delay and jitter. The average relative error of the estimate does not exceed 4. 1%.
25298 实验也对比了传统最短路径路由算法与基于该预测模型给出的最优路由方案下的端到端性能，相比传统最短路径路由算法，平均延迟和平均抖动分别降低了19. 8% 和 33. 52% ，最大延迟和最大抖动降低了 36. 18% 和 35. 45% 。 The experiment also compares the end-to-end performance of the traditional shortest path routing algorithm with the optimal routing scheme based on the prediction model proposed in this paper. Compared to the traditional shortest path routing algorithm, the average delay and average jitter are reduced by 19. 8% and 33. 52% , and the maximum delay and maximum jitter are reducedby 36. 18% and 35. 45% ．
25299 针对长短时记忆神经网络(Long Short-Term Memory，LSTM)模型计算开销大、冗余计算较多的问题，本文提出一种利用输入数据稀疏性的 LSTM 加速器设计方案。 Aiming at the problem of high computational overhead and redundancy in LSTM model，this paper proposed a design method of LSTM accelerator based on data sparsity．
25300 本方案基于 Delta 网络算法，对输入序列的稀疏性进行构建，在避免数据不规则加载的前提下，对冗余矩阵向量乘法运算进行过滤; This scheme uses Delta network algorithm to mine the sparsity of input data, and filters the multiplication of redundant matrix vectors without irregular loading of data.
25301 针对矩阵向量乘法计算模式进行建模，寻找最高效的并行阵列计算架构设计。 The calculation mode of matrix-vector multiplication is modeled to find the most efficient parallel array computing architecture design scheme．
25302 在 MNIST 标准数据集上的实验表明，当 Delta 网络算法的过滤门限不超过 0. 5 时，LSTM 神经网络算法检测准确率不变，计算性能提高了 21. 53 倍。 The experimental results on MINIST dataset show that when the filter threshold of Delta network algorithm is less than 0. 5, the detection accuracy of LSTM neural network algorithm remains unchanged, and the computational performanceis improved by 21. 53 times．
25303 针对现有基于校验关系的 Turbo 码交织器识别算法在低信噪比(Signal-to-Noise Ｒatio，SNＲ)时适应性较弱以及识别过程中容易出现差错传播的问题，为了进一步提升交织识别的容错性与实时性，在基于最大序列相关识别算法的基础上加入迭代译码和小范围遍历两种纠错算法，更加充分的利用了接收数据的同时增强了抗误码能力。 The existing algorithms based on parity-check relationship in the recognition of interleaver for Turbo codeshave weak adaptability at low signal-to-noise ratio (SNＲ)and are prone to "error propagation" in the recognition process. In order to further improve the fault tolerance and real time of interleaving recognition, we propose two error correction algorithms based on the maximum sequen cecorrelation recognition algorithm, including iterative decoding and small traversal, which make full use of the received data and enhance the error resilience．
25304 仿真结果表明，加入纠错算法使得原有算法的交织识别性能明显改善，相同条件下完成全部交织参数识别所需数据量仅需原有的 1 /3;在相同数据量的条件下实现全部交织参数识别时的 SNＲ 增益大于 2dB。 Simulation results show that the performance of the improved algorithm is obviously better than the original algorithm by adding error correction algorithm, and it only needs 1 /3 of the data to complete the identification of all the interleaved parameters under the same conditions and has more than 2dB gain to achieve the identification of all the interleaved parameters under the condition of the same amount of data．