Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2020-09-28 Cooperative journals: 《计算机应用研究》
Abstract: Aiming at the traditional models parallel method for accelerating convolution neural network (CNN) tend to have low utilization, this paper proposed PipeCNN, a method that accelerates CNN with software pipeline. Firstly, this paper studied the forward propagation and back propagation, and then explored data correlation during training. Secondly, it parallelized CNN with the support of software pipeline, and then analyzed two feasible gradient updating methods in PipeCNN. Finally, it used circular queue to realize communication between two layers and then proposed a task allocation algorithm to divide CNN into working parts. Experiments showed that the method can obtain good speedup and utilization while ensuring the accuracy of the model. It showed that software pipeline can effectively solve the problem of low utilization in model parallel, and accelerate the training of CNN.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2019-01-03 Cooperative journals: 《计算机应用研究》
Abstract: In view of the low performance of traditional TCP protocol in transmitting large files in the current network environment, this paper studied and optimized the slow start algorithm of TCP transport protocol. According to the problems of standard slow-start algorithm and the performance characteristics of high-speed network and large file transmission, an improved slow-start algorithm with network state awareness is proposed in this paper. The improved algorithm mainly optimizes the growth strategy of TCP congestion window, which monitors the round-trip time (RTT) of TCP packets in real-time during the transmission process, and updates the congestion window with different window growth methods according to the changes of RTT. It deployed the improved algorithm in the Linux network module and tested it in the simulated network environment and the actual network. Experimental results show that the improved algorithm can keep the sending window at a higher level, and the actual data transmission rate and throughput have been significantly improved.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-12-13 Cooperative journals: 《计算机应用研究》
Abstract: Aiming at the problem that the feature cascading method for face feature extraction based on Gabor wavelet amplitude and phase makes the feature vector dimension higher, this paper proposed an improved Gabor wavelet transform feature extraction algorithm. The algorithm calculated local amplitude and local phase features, and enhanced the local correlation of each pixel. Then the algorithm weighted fusion for amplitude and phase features by weighting coefficients which experiment selected. Experimental results show that the proposed algorithm reduces the dimension of the feature vector and improves the final face recognition rate compared with the pre improved algorithm.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-05-18 Cooperative journals: 《计算机应用研究》
Abstract: Shilling attack is one of serious challenges which recommender systems are facing. Malicious users can easily insert well-designed ratings into recommender systems to affect recommendation results and decrease user experiences because of the openness of recommender systems. This article proposed a robust shilling-attack-tolerant personalized recommendation (SATPR) algorithm based on attribution facilitated matrix completion with structural noise technology, regarded the ratings of attack users in the rating matrix as structural row noise and modeled them with L2,1-norm. This article also introduced attributive characters of users and items to improve the accuracy of detection of shilling-attack. Experimental results showed that SATPR algorithm achieved more accurate results of personalized rating prediction than traditional recommendation algorithms under shilling attacks.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-04-19 Cooperative journals: 《计算机应用研究》
Abstract: There are lots of drawbacks in existing attribute-based encryption with keyword search scheme, such as privacy disclosure issues and when the authorized user is offline how effectively delegate decrypt and search right to other people. In order to solve these problems, this paper proposed an attribute-based ciphertext search scheme with hidden access structures, which support proxy re-encryption function by combing attribute-based encryption with hidden access structures and proxy re-encryption technology. This scheme not only effectively solves the problems mentioned above, but also supports the update of keywords. Finally, under the random oracle model, the researchers proved the security of the scheme based on DL (D-linear) and q-BDHE (decisional q-parallel bilinear Diffie-Hellman exponent) hypothesis.