Collaborative Filtering Matches Decision Templates: A Practical Approach to Estimate Predictions

This paper deals with insulating part quality control according to image analysis. The studied insulating parts are mainly composed of glass fibres and their orientation is directly correlated to the quality of the parts. This complex phenomenon is analyzed by means of 3D-tomographic images which give a huge set of raw data. Relevant features were extracted by several classification approaches to detect interesting regions defined by experts. The paper focuses on a fusion system based on decision templates to aggregate the previously obtained image classifications. The initial decision templates method proposed by L. Kuncheva is adapted for the image analysis concerned: (1) a solution is proposed to take into account classes of rejects for which there are no reference regions to learn the corresponding decision templates and (2) neighboring pixels are considered inside the fusion process of the decision templates. The fusion approach is then applied to part quality control. Results are assessed by means of the confusion matrix and accuracy measures show the great improvement in region detection brought by the fusion approach according to each input classification.

Extended decision template presentation for combining classifiers

Using measures of similarity and inclusion for multiple classifier fusion by decision templates

Blind universal steganalysis has been the choice of Steganalysers owing to it’s capability to detect stego images without any prior information about the embedding method. Universal steganalysis is a two class optimization problem and the detecting efficiency depends on the feature set chosen from the stego and clean images. Though extracting all possible features of an image may lead to more efficiency the classification suffers due to large dimension of feature set. To overcome the problem of dimensionality appropriate feature reduction techniques need to be employed. This paper presents a blind universal image steganalysis technique that extracts the noise models of adjacent pixels of an image. The exact model construction involves the formation of four dimensional co-occurrence matrices of the quantised and truncated noise residues. From the 106 sub models 34,671 features have been extracted and further reduced by a novel unsupervised optimization technique to identify the most appropriate features for classification. The classifiers implemented include Support Vector Machines (SVM), Multi Layer Perceptron (MLP) and three fusion classifiers based on Bayes, Decision Template and Dempster Schafer fusion schemes. It has been identified that MLP performs better than SVM but is not superior to fusion classifiers. Comparing all the classifiers, Decision Template based fusion method gives the best classification accuracy (99.25%). Thus the proposed unsupervised optimization method combined with Decision Template fusion classification scheme provides the best classification of stego and clear images as compared to the existing research work.

In this paper, a new classifier fusion method is introduced based on a decision template structure as an extension to Dempster Shafer method. It employs multilayer neural networks as base classifiers. The idea relies on the fact that in a multilayer neural network, behavior of each layer can be a guide for modeling decision-making process. The new decision template based method constructs decision template for each layer of the neural networks including all hidden layers such that a complete model of the base classifiers decision making process is built. In the combiner part, a new strategy based on extension to Dempster Shafer method is introduced. Efficiency of this method is compared with some known benchmark datasets.

Multiclass pattern recognition problems (K > 2) can be decomposed by a tree-structured approach. It constructs an ensemble of K -1 individually trained binary classifiers whose predictions are combined to classify unseen instances. A key factor for an effective ensemble is how to combine its member outputs to give the final decision. Although there are various methods to build the tree structure and to solve the underlying binary problems, there is not much work to develop new combination methods that can best combine these intermediate results. We present here a trainable fusion method that integrates statistical information about the individual outputs (clustered decision templates) into a Radial Basis Function (RBF) network. We compare our model with the decision templates combiner and the existing nontrainable tree ensemble fusion methods: classical decision tree-like approach, product of the unique path and Dempster-Shafer evidence theory based method.

Mixed pixels of hyperspectral image own very high sparsity if they are linearly represented by endmembers of aprior spectral library. Hence the sparse regression framework has been introduced to solve the linear spectral unmixing problem and produced sparse fractional abundances. But because of the high coherence of spectral library and noise disturbance, the solution could not be as sparse as real. Structure sparsity is introduced to the sparse regression based unmixing method to mitigate the disturbances and make the solution sparser and more accurate. This paper gives an overview of the structure feature of hyperspectral image such as spatial smoothness, collaborative sparsity and group structure of spectral library as well as corresponding structure sparsity based hyperspectral unmixing methods. Furthermore some ideas of future work on structure sparsity in the unmixing of hyperspectral image are proposed at last.

Bio inspired optimization for universal spatial image steganalysis

The performance accuracy of JPEG steganalysis depends on the best features extracted from the images. This demands extraction of all possible features that undergo changes during embedding. The computational complexity due to such large number of features necessitates feature set optimization. Existing research in JPEG image steganalysis tend to extract rich feature sets and reduce them by statistical feature reduction techniques. Compared to these techniques, genetic algorithm based optimization techniques are more promising as they converge to global minima. The objective of this paper is to implement genetic based optimization to reduce the high dimensional image features and hence obtain improved classification accuracy. The method implemented includes the extraction of image features in terms of co-occurrence matrices of the differences of all possible Discrete Cosine Transform (DCT) coefficients to give 200 × 23,230 features. These features are optimized by a nature inspired meta-heuristic, Ant Lion Optimization (ALO) which considers the features as ants that move in random space. The fitness function for the antlion to hunt the ants is proportional to the traps built by the antlion. The proposed steganalyser has been tested for classification accuracies with different payloads. The classifiers implemented include Support Vector Machines (SVM), Multi Layer Perceptron (MLP) and fusion classifiers based on Bayes, Decision template and Dempster Schafer data fusion schemes. The results show that highest average classification accuracy has been obtained for Bayes fusion classifier followed by Dempster Schafer fusion classifier. It has been noted that the performance of fusion classifiers is better compared to individual classifiers. Thus the proposed method gives better classification accuracy for JPEG steganalysis than existing methods.

In this paper, a nonlinear model for medium-term hydro-thermal scheduling problems with transmission constraints is presented. A nonlinear formulation of hydro power production functions of discharge and storage is used for proper representation of head variation in cascaded reservoirs. Transmission networks are formulated as a direct-current power flow model with support to load attainment over multiple levels representing peak and off-peak hours, enabling varying degrees of both network topology and load balance nodes representation over time stages. Sparse matrix structures that arise from the mathematical formulation allow fast decoupled computation of search directions in a primal-dual interior-point framework. Computational tests with very large case studies are presented. Analysis of numerical algorithm convergence is provided along with insightful optimal solution interpretative commentary.

The end of Dennard scaling (i.e., the ability to shrink the feature size of integrated circuits while maintaining a constant power density) has now placed energy as a primary design principle in par with performance, all the way from the hardware to the application software. Along this line, optimizing the performance-energy balance of the 7/13 dwarfs, introduced by UC Berkeley in 2006, represents a challenge with a potential tremendous impact on a vast number of scientific applications. However, without a careful modeling and a subsequent understanding of the performance-energy interaction, the optimization process of these kernels is doomed to fail. In this paper we investigate the performance-power-energy characterization of the sparse matrix-vector product (SpMV), a challenging kernel due to its indirect and irregular memory access pattern, which constitutes the key ingredient of the sparse linear algebra dwarf. In the first part of our analysis we identify a reduced set of critical features (based on statistics about the sparse matrix structure) which impact the performance, power, and energy consumption of a baseline implementation of SpMV. We then generate a small synthetic sparse benchmark collection (the training set) that we use to build (i) a general classification of sparse matrices and (ii) a model to accurately predict performance and energy consumption of any SpMV. Both tools are based on the features (parameters) emerged from the first part of our study, and they are validated using the entire University of Florida Matrix Collection, run on two high-end multithreaded architectures.

Many block ciphers and hash functions use MDS matrices because of their optimal branch number. On the other hand, MDS matrices generally have a high implementation cost, which makes them unsuitable for lightweight cryptographic primitives. In this direction, several sparse matrix structures like companion, GFS, and DSI matrices are proposed to construct recursive MDS matrices. The key benefit of these matrices is their low fixed XOR, and the diffusion layer can be made by recursively executing the implementation of the matrices, which takes a few clock cycles. In this paper, we propose a new class of sparse matrices called Diagonal-like sparse (DLS) matrices and the DSI matrix is a particular type of DLS matrix. We prove that for an n-MDS DLS matrix of order n, the fixed XOR (say \(\mathcal {K}\)) should be at least equal to the \(\left\lceil \frac{n}{2}\right\rceil \). We also show that an n-MDS DLS matrix over \(\mathbb {F}_{2^r}\) with \(\mathcal {K}=\left\lceil \frac{n}{2}\right\rceil \) is a permutation similar to some n-MDS sparse DSI matrix. We propose another type of sparse matrices called generalized DLS (GDLS) matrices. Next, we introduce some lightweight recursive MDS matrices of orders 4, 5, 6, and 7, using GDLS matrices, that can be implemented with 22, 30, 31, and 45 XORs over \(\mathbb {F}_{2^8}\), respectively. The results match the best known lightweight recursive MDS matrices of orders 4 and 6 and beat the best known matrices of orders 5 and 7. Also, the proposed 4-MDS GDLS matrix over \(\mathbb {F}_{2^4}\) has a XOR count of 10, which meets the best known result.KeywordsDiffusion layerMDS matrixPermutation matrixXOR count

Despite the high level of sophistication of machine tools, they are still subject to geometric errors arising from various sources, which directly affect the dimensional accuracy of machined parts. Given the limitations of traditional calibration methods, such as high cost, time consumption, and complexity, this study proposes and applies an alternative method based on error separation for geometric evaluation in a three-axis CNC machining center, demonstrating its practical feasibility. Using a slender 1045 steel artifact and a pair of LVDT sensors with a resolution of 0.0001 mm, measurements were performed in six different setups, with the instruments and artifact fixed to the machine structure. The mathematical model developed has a sparse matrix structure and was solved numerically using the LSQR algorithm, reaching satisfactory convergence in 130 iterations. The straightness profiles obtained through the proposed method were compared to measurements taken with a CMM, showing agreement. The analysis of angular and linear errors allowed the identification of a critical region at 320 mm along the y-axis, where the largest deviations occur, indicating points for applying numerical compensation or mechanical adjustments. The method stood out for enabling the mathematical removal of errors and reconstruction of the artifact's profile, as well as for its ability to locate and quantify geometric errors in regions crucial for defining the cutting height of the tool. The modeling and algorithm used allow for precise results with low computational cost. Thus, the study presents significant potential for calibration and predictive maintenance of machine tools, providing technical information for decisions on error compensation and quality improvement in the machining process.

In orthogonal frequency division multiplexing (OFDM) systems, when the length of the cyclic prefix (CP) is shorter than the channel length, the orthogonality between sub-channels is lost because of the intersymbol interference (ISI) and interchannel interference (ICI). In this case, the one-tap frequency domain equalizer can't be used any more. A time-domain equalizer (TEQ) is usually used in the receiver to reduce the duration of the overall response of the transmission system, and therefore minimize the ISI and ICI. However, the optimum design of TEQ turns out to be a very difficult task. In this paper, we propose a frequency domain equalizer (FEQ) for OFDM systems with insufficient CP, by making use of the presence of null side sub-carriers and the redundancy of CP. The equalizer has a sparse matrix structure and thus a low computational complexity. Theoretical analysis and simulation results show that it can efficiently remove ISI and ICI caused by insufficient CP and recover the transmitted data. Moreover, we derive the condition for the existence and uniqueness of FEQ, i.e., the combined length of CP and null sub-carriers is not shorter than the channel order. This means that the insufficiency of CP in time domain can be compensated by the redundancy of the null side sub-carriers in frequency domain.

In this paper, we present a new sparse matrix data format that leads to improved memory coalescing and more efficient sparse matrix-vector multiplication for a wide range of problems on high-throughput architectures such as a GPU. The sparse matrix structure is constructed by sorting the rows based on the row length (defined as the number of non-zero elements in a matrix row) followed by a partition into two ranges, short rows and long rows. Based on this partition, the matrix entries are then transformed into ELLPACK or vectorized compressed sparse row format. In addition, the number of threads are adaptively selected by their row length, in order to balance the workload for each graphics processing unit thread. Several computational experiments are presented to support this approach and the results suggest a notable improvement over a wide range of matrix structures.