In order to solve the problem of insufficient available training data in the classification task of breast mass and calcification, a multi-view model based on secondary transfer learning was proposed combining with imaging characteristics of mammogram. Firstly, CBIS-DDSM (Curated Breast Imaging Subset of Digital Database for Screening Mammography) was used to construct the breast local tissue section dataset for the pre-training of the backbone network, and the domain adaptation learning of the backbone network was completed, so the backbone network had the essential ability of capturing pathological features. Then, the backbone network was secondarily transferred to the multi-view model and was fine-tuned based on the dataset of Mianyang Central Hospital. At the same time, the number of positive samples in the training was increased by CBIS-DDSM to improve the generalization ability of the network. The experimental results show that the domain adaption learning and data augmentation strategy improves the performance criteria by 17% averagely and achieves 94% and 90% AUC (Area Under Curve) values for mass and calcification respectively.
Aiming at the concept drift problem, a classification learning model with the characteristics of data changing progressively over time was constructed, and a Gradual Multiple Kenerl Learning method (G-MKL) based on Gradual Support Vector Machine (G-SVM) was proposed. In this method, with Support Vector Machine (SVM) used as the basic classifier, multi-interval sub-classifier coupling training was carried out and the incremental method of constraining sub-classifier was used to adapt the model to the gradual change of data. Finally, multiple kernels were integrated into SVM solution framework in a linear combination manner. This method integrated the advantages of different kernel functions and greatly improved the adaptability and validity of the model. Finally, the comparison experiments between the proposed algorithm and several classical algorithms were carried out on the simulated and real datasets with gradual characteristics, verifying the effectiveness of the proposed algorithm in dealing with non-stationary data problems.
In order to solve the problems of keyword extraction and project keyword lexicon establishment of technological projects in professional fields, an algorithm for building the lexicon based on semantic relation and co-occurrence matrix was proposed. On the basis of conventional keyword extraction research based on co-occurrence matrix, the algorithm considered several advanced factors such as the location, property and Inverse Document Frequency (IDF) index of the keywords to improve the traditional approach. Meanwhile, a method was given for the establishment of keyword semantic network using co-occurrence matrix and hot keyword identification through computing the similarity with semantic base vector. At last, 882 project experiment documents in power field were used to perform the simulation. And the experimental results show that the proposed algorithm can effectively extract the keywords for the technological projects, establish the keyword correlation network, and has better performance in precision, recall rate and F1-score than the keyword extraction algorithm of Chinese text based on multi-feature fusion.
To solve the problem that current Decentralized Information Flow Control (DIFC) systems are unable to monitor the integration of host and network sensitive data effectively, a new design framework of DIFC system based on Software Defined Network (SDN), called S-DIFC, was proposed. Firstly, this framework used DIFC modules to monitor files and processes in host plane with fine granularity. Moreover, label mapping modules were used to block network communication and insert sensitive data labels into network flow. Meanwhile the multi-level access control of the flow with security label was implemented with SDN's controller in network plane. Finally, S-DIFC recovered security labels carried by sensitive data in DIFC system on target host. The experimental results show S-DIFC influences host with CPU performance decrease within 10% and memory performance decrease within 1.3%. Compared to Dstar system with extra time-delay more than 15 seconds, S-DIFC mitigates communication overhead of distributed network control system effectively. This framework can meet the sensitive data security requirements of next generation network. In addition, the distributed method can enhance the flexibility of monitor system.
Oblivious transfer plays an important role in the field of cryptography. A provably secure k-out-of-n oblivious transfer scheme was analyzed in this paper. This scheme was based on a novel method and was efficient in computation and communication. However, it was found not secure at all after deep analysis. The main fault is that the receiver can easily acquire all the secret messages sent by sender. Thus it does not satisfy the secure requirement of oblivious transfer. Finally, by adding a random number the fault of the scheme was fixed. The improved k-out-of-n oblivious transfer scheme keeps the same communicational overhead and computational overhead as the original one. The security of the improved scheme is also based on Decisional Diffie-Hellman (DDH) assumption.
Concerning the large-scale concurrent video stream scheduling problem of low resource utilization and load imbalance under cloud environment, a Video-on-Demand (VOD) scheduling policy based on Ant Colony Optimization (ACO) algorithm named VodAco was proposed. The correlation of video stream expected performance and server idle performance was analyzed, and a mathematical model was built based on the definition of comprehensive matching degree, then ACO method was adopted to hunt the best scheduling schemes. The contrast experiments with Round Robin (RR) and greedy schemes were tested on CloudSim. The experimental results show that the proposed policy has more obvious advantages in task completion time, platform resources occupancy and node load balancing performance.
由于缺少结构化的表示,基于内容的图像分类存在一定的问题,据此提出了一种基于迭 代神经网络的自然图像表示和分类的方法。利用Berkeley分割算法将图像分割成不同的区域,采用 基于人工的多叉树或基于邻接区域的二叉树的方法进行区域合并,同时提取区域统计特征,得到图像 的树型结构表示。根据BPTS算法对网络进行训练,训练好的网络就具备了图像分类的功能。实验 结果表明,基于迭代神经网络的结构表示和分类方法具有很强的结构学习能力,同时人工生成的多叉 树涵盖更多的语义信息且能得到较好的分类结果。
This paper analyzed the principle of collecting information of network topology with SNMP and ICMP, the faults of these means and some advice were discussed. Then, a new RT algorithm of remote network topology by synthetizing many probing techniques was presented. The algorithm not only put emphasis on the collection of network devices infomation,but also gave prominence to the analysis of these infomation. Finaly, the integrality and veracity of the result of this new network topology discovery algorithm were proved by compared the results of different way,all based on an emulating network environment.