recurrent neural network language models have solved the problems of data sparseness anddimensionality disaster which exist in traditional N-gram models. rNNLMs have recently demonstrated state-of-the-art performance...
详细信息
ISBN:
(纸本)9781479932801
recurrent neural network language models have solved the problems of data sparseness anddimensionality disaster which exist in traditional N-gram models. rNNLMs have recently demonstrated state-of-the-art performance in speech recognition, machine translation and other tasks. In this paper, we improve the model performance by providing contextual word vectors in association with rNNLMs. This method can reinforce the ability of learning long-distance information using vectors training from Skip-gram model. The experimental results show that the proposed method can improve the perplexity performance significantly on Penn Treebank data. And we further apply the models to speech recognition task on the Wall Street Journal corpora, where we achieve obvious improvements in word-error-rate.
This paper is concerned with the iteration identification algorithm for Hammerstein model with complex-valued input for the fact that the existing algorithms are not valid for complex input. Based on the stochastic gr...
详细信息
ISBN:
(纸本)9781510804166
This paper is concerned with the iteration identification algorithm for Hammerstein model with complex-valued input for the fact that the existing algorithms are not valid for complex input. Based on the stochastic gradient algorithm, the extended stochastic gradient algorithm is proposed by defining new cost function for complex input. The extended hierarchical multi-innovation stochastic gradient algorithm is proposed by introducing multi-innovation identification theory and hierarchical principle to the extended stochastic gradient algorithm. Experimental simulations show that the extended hierarchical multi-innovation stochastic gradient algorithm has better performance than the extended stochastic gradient algorithm at the expense of computational complexity.
In this paper,a novel UWB communication system structure is *** transmitted signal uses OOK modulation and chirp spread *** received signal first goes through a dechirp pulse compression process,and then is processed ...
详细信息
In this paper,a novel UWB communication system structure is *** transmitted signal uses OOK modulation and chirp spread *** received signal first goes through a dechirp pulse compression process,and then is processed with a two-level sliding correlation algorithm for coarse timing synchronization and fine timing *** the SNr estimation,the signal is demodulated by an energy detection *** integratedsystem level simulation model is established,and the performance of this system is evaluated over the AWGN channel,IEEE 802.15.3a CM1 and CM4 *** theoretical analysis and simulation results show that this UWB communication system can effectively reduce the sampling rate and signal processing speed at the receiver,and it is more suitable for long-distance and low-rate UWB communications with high spreading gain.
Today's Internet hosts a wide variety of network objects (namely, netOs) like contents, resources and functionalities and so on, which are often decomposed anddistributed over the Internet to provide and support ...
详细信息
Today's Internet hosts a wide variety of network objects (namely, netOs) like contents, resources and functionalities and so on, which are often decomposed anddistributed over the Internet to provide and support network-based services and applications conveniently. However, it is a big challenge to efficient management due to the huge number of existing netOs. Based on our investigations, most existing works that organizing those netOs in the computer network efficiently formulates network utility maximization problems (or network cost minimization). Although some works are effective in the corresponding aspects, the Internet still suffers from inefficiency and ossification. Instead of solving a concrete problem, we make the first attempt to model the numerical evolution of netOs (namely, EvoNetO) to uncover the hidden order of netOs. We prove that the problem of minimizing network cost is equivalent to fastest decay rate or slowest growth rate in the population dynamics. And then we find that the netOs distributed over the network will go to two extreme, i.e., r-strategists and K-strategists under the goal of minimizing network cost for netO placements. This phenomenon exists widely in the computer network.
Ternary Content-Addressable Memories(TCAMs) have become the industrial standard for high-throughput packet *** comparison against all of its entries in parallel within a single memory access makes TCAMs power *** this...
详细信息
ISBN:
(纸本)9781467349970
Ternary Content-Addressable Memories(TCAMs) have become the industrial standard for high-throughput packet *** comparison against all of its entries in parallel within a single memory access makes TCAMs power *** this paper,we propose a new algorithm called TSP-Pr(Tri-statebased Partition for Powerreduction).In TSP-Pr,identification bits are dynamically selected to partition a classifier into one general subset and many dedicated *** packet arrives,only one dedicated subset and the general subset are activated for parallel *** makes a great powerreduction since the main component of power consumption of TCAMs is proportional to the number of searched *** experimentation shows that the TSP-Pr algorithm proposed in this paper can highly reduce powerdissipation with a little penalty of storage overhead.
In order to solve the problem of excessive IP(Intellectual Property) usage when calculating large-size squares by using embedded multipliers in FPGA directly,this paper presents an efficient method for square *** prop...
详细信息
ISBN:
(纸本)9781467349994
In order to solve the problem of excessive IP(Intellectual Property) usage when calculating large-size squares by using embedded multipliers in FPGA directly,this paper presents an efficient method for square *** proposed methodreduces the width of operands by a simple iterative logic circuit,then realizes square calculation combine with the embedded *** a 32-bit square calculation,experimental and simulation results show that the method can get goodresults only by using one dSP48E with a small amount of logic resources.
Because of the problems of the existing transmission mode for streaming media system that could not guarantee the reception quality of the data while keeping the high distribution efficiency,a new adaptive transmissio...
详细信息
ISBN:
(纸本)9781467349970
Because of the problems of the existing transmission mode for streaming media system that could not guarantee the reception quality of the data while keeping the high distribution efficiency,a new adaptive transmission mode selection algorithm is proposed in this *** the algorithm,with a new node state model which divides the node state into data receive quality and consolidated state,the client node selects the transmission mode adaptively based on the existing data buffer model,while the server nodes choose the super nodes to push the data *** simulation results show that this algorithm enhances the distribution efficiency,while the system scalability is improved under the control of the server load.
Fault model is the basis of research on fault-tolerant routing algorithm for Network on Chip(NoC).Since the former fine-grained fault model could not distinguish link faults from channel faults effectively,a much more...
详细信息
ISBN:
(纸本)9781467349994
Fault model is the basis of research on fault-tolerant routing algorithm for Network on Chip(NoC).Since the former fine-grained fault model could not distinguish link faults from channel faults effectively,a much more fine-grained fault model is put forward,and a new fault-tolerant routing algorithm named VFFrA-Br(very fine-grained fault-tolerant routing algorithm based on bufferreuse) is proposed in this paper,to improve the utilization rate of the NoC *** algorithm adopts the transparent transmission mechanism based on bufferreuse(TTBr) to tolerate internal channel faults of *** and synthesis results demonstrate that our algorithm performs better on average throughput,average latency and average hop counts of NoC while keeping a low onchip resources overhead.
The network application of flow level becomes increasingly wide,to analyze network flow completely we need fair *** previous algorithm of S GS(S ketch Guided S ampling) realizes fair sampling by that sampling rate is ...
详细信息
ISBN:
(纸本)9781467349970
The network application of flow level becomes increasingly wide,to analyze network flow completely we need fair *** previous algorithm of S GS(S ketch Guided S ampling) realizes fair sampling by that sampling rate is in inverse ratio to traffic,traffic counting of which uses hash *** this paper,for the defects of S GS flow counting and the previous counting Bloom Filters,a more efficient fair sampling algorithm calleddCFS(dynamic Count Fair S ampling) is *** algorithm of dCFS applies dCF(dynamic Count Filter) to count traffic,which is faster more space efficient than S GS,and it can be easily implemented in hardware.
In order to improve the location accuracy in "urban canyons" and indoor environments,a new algorithm based on MIMO-OFdM technique is ***,the errorrate of the subcarriers in the OFdM signal is reduced by the...
详细信息
ISBN:
(纸本)9781467349994
In order to improve the location accuracy in "urban canyons" and indoor environments,a new algorithm based on MIMO-OFdM technique is ***,the errorrate of the subcarriers in the OFdM signal is reduced by the transmitterdiversity and maximum ratio combining in MIMO;and then the operation of cross-correlation between the received signal andreference signal is processed to obtain the time delay estimation;finally,we use the time delay to realize the *** simulation results show that the location accuracy in proposed algorithm is improved about 25 m compared to the reference algorithm in the cumulative distribution probability of 90%.
暂无评论