To solve the problem that Explicit Shape Regression (ESR) has low precision in face alignment, an improved explicit shape regression for face alignment algorithm was proposed. Firstly, in order to get a more accurate initial shape, three-point face shape was used as an initial shape mapping standard to replace face rectangle. Then, pixel block feature was used against illumination variations instead of pixel feature, which improved the algorithm robustness. Finally, instead of average method, the accuracy of algorithm was further improved by multiple hypothesis fusion strategy which merged multiple estimations. Compared with explicit shape regression algorithm, the simulation experimental results show that the accuracy is improved by 7.96%, 5.36% and 1.94% respectively by using the proposed algorithm on LFPW, HELEN and 300-W face datasets.

All the previous certificate-based Proxy Re-Encryption (PRE) schemes are based on the computationally-heavy bilinear pairings, and thus have low computation efficiency. To solve this problem, a certificate-based proxy re-encryption scheme without relying on the bilinear pairings was proposed over the elliptic curve group. Under the hardness assumption of the Computational Diffie-Hellman (CDH) problem, the proposed scheme was formally proven to be indistinguishable against adaptively chosen-ciphertext attacks in the random oracle model. Due to avoiding the time-consuming bilinear pairing operations, the proposed scheme significantly reduced the computation cost. Compared with the previous certificate-based proxy re-encryption schemes with bilinear pairings, the analysis shows that the proposed scheme has obvious advantages in both the computation efficiency and the communication cost, and the scheme is more suitable for the computation-constrained and bandwidth-limited applications.

Aiming at the problems that the traditional Support Vector Machine (SVM) classifier is sensitive to outliers and has the large number of Support Vectors (SV) and the parameter of its separating hyperplane is not sparse, the Truncated hinge loss SVM with Smoothly Clipped Absolute Deviation (SCAD) penalty (SCAD-TSVM) was put forward and was used for constructing the financial early-warning model. At the same time, an iterative updating algorithm was proposed to solve the SCAD-TSVM model. Experiments were implemented on the financial data of A-share manufacturing listed companies of the Shanghai and Shenzhen stock markets. Compared to the T-2 and T-3 models constructed by SVM with L1 norm penalty (L1-SVM), SVM with SCAD penalty (SCAD-SVM) and Truncated hinge loss SVM (TSVM), the T-2 and T-3 model constructed by the SCAD-TSVM had the best sparseness and the highest accuracy of prediction, and its average accuracies of prediction with different number of training samples were higher than those of the L1-SVM, SCAD-SVM and TSVM algorithms.

The kernel function methods can discover the nonlinear distribution rules among the images of high precision prints. And the mining capacity is decided by the kernel function and its parameters. Selecting the kernel function is imminent to the development and application in kernel function theory. Based on the intelligent detection of prints, a new learning kernel method based on the optimization was presented for the industry of high precision printing to make the kernel function method to achieve optimal performance. Unlike the traditional calculation method, the kernel's parameter was continuously changing in kernel space, which meant that the learning scope expanded one dimension. The experimental results show that the iterative algorithm based on the theoretical analysis only needs ten iterations to get the statistical optimal kernel function and its parameters, and the recovery error of the kernel function is statistically minimum.

With the continuing increase of users and kinds of resources, the problem of rating matrix's sparsity is becoming more and more prominent, which seriously affects the quality of the recommendation system. Singular Value Decomposition (SVD) is a dimension reduction method, and Symbolic Data Analysis (SDA) is a new analytical approach to processing mass data. This paper proposed a new collaborative filtering recommendation algorithm which combines SVD with SDA. The experimental results based on EachMovie database set indicate that the proposed method is significantly better than traditional general recommendation algorithm when the data is particularly sparse.

With the advancement of Graphics Processing Unit (GPU) and the creation of its new feature of programmability, many algorithms have been successfully transferred to GPU. LU decomposition and Laplace algorithms are the core in scientific computation, but computation is usually too large; therefore, a speedup method was proposed. The implementation was based on Nvidia's GPU which supported Compute Unified Device Architecture (CUDA). Dividing tasks on CPU and GPU, using shared memory on GPU to increase the speed of data access, eliminating the branch in GPU program and stripping the matrix were used to speed up the algorithms. The experimental results show that with the size of matrix increasing, the algorithm based on GPU has a good speedup compared with the algorithm based on CPU.

To address the issues of resource availability, limited simulation scale, etc. in simulation system, an distributed simulation platform, which was based on grid technology and HLA, was proposed in this paper. The idea of separating federate into simulation client and simulation service provider was discussed. This platform was implemented. The result shows introducing the grid technology into HLA simulation provide a way to improve resource availability and to enlarge the scale of simulation system.

A new search angine ranking algorithm named CategoryRank based on categorization was proposed, in order to yield more accurate search results. With this new algorithm, the notion of importance can be captured more accurately with respect to a particular category. For this, the link graph was first analyzed and computed based on the category difference between two web pages of this link to better reflect user behavior in surfing. Secondly, the category information was applied into each web page to distinguish the importance of this page for different kind of users. Finally, the offline model and online model of this algorithm were combined to detail the application in ranking of search engine.

Scheduling risk is one of the most critical risks of software project. The scheduling risk management based on task critical chain was discussed in this paper. Each task’s duration in ideal working environment was estimated, personnel resource constraints were taken into account, and the project’s critical chain was built. After the risk analysis for each task, the project buffer was established for the critical chain and feeding buffer was established for non-critical chain. The buffer area was treated as resort of risk control and management.