Search

Compression is desirable for network applications as it saves bandwidth; however, when data is compressed before being encrypted, the amount of compression leaks information about the amount of redundancy in the plaintext. This side channel has led to successful CRIME and BREACH attacks on web traffic protected by the Transport Layer Security (TLS) protocol. The general guidance in light of these attacks has been to disable compression, preserving confidentiality but sacrificing bandwidth. In this paper, we examine two techniques—heuristic separation of secrets and fixed-dictionary compression—for enabling compression while protecting high-value secrets, such as cookies, from attack. We model the security offered by these techniques and report on the amount of compressibility that they can achieve.

Inter-picture coding schemes are operating equivalent to intra-picture coding schemes presented in the previous chapter, as far as coding of 2D picture information is concerned. Similarity between (temporally or spatially) adjacent pictures can now be exploited by inter-picture compression, wherein techniques of displacement compensation play a key role. Firstly, high redundancy between subsequent pictures in a video sequence exists along the motion trajectory; furthermore, coding errors are less visible if they are consistent with the motion of the scene content. This chapter also introduces more simple schemes without motion compensation and shows their limitations.

Regarding text region as region of interest (ROI) and assigning higher bit budget to ROIs than rest regions, ROI-based text image compression can provide both higher quality for text regions and higher compression ratio for an entire text image. JPEG2000 is a high performance image compression standard for common images but has no special optimization for text images. After image characteristic analysis, this paper proposed a natural scene text image compression method based on JPEG2000 ROI coding. ROI coding parameters are optimized for text regions. In this stage, the redundancy analysis is used to measure compression capability of different regions and scale factors in ROI coding are adjusted adaptively. With the specially designed optimization, the proposed method shows practicality in real applications. The experiment results show the improvement of compression performance which verifies the optimization effectiveness.

Feature sets in many domains often contain many irrelevant and redundant features, both of which have a negative effect on the performance and complexity of agents that use the data [ 9 ]. Supervised feature selection aims to overcome this problem by selecting features that are highly related to the class labels, yet unrelated to each other. One proposed technique to select good features with few inter-dependencies is minimal Redundancy Maximal Relevance mRMR [ 12 ], but this can be impractical with large feature sets. In many situations, features are extracted from signal data such as vehicle telemetry, medical sensors, or financial time-series, and it is possible for feature redundancies to exist both between features extracted from the same signal intra-signal, and between features extracted from different signals inter-signal. We propose a two stage selection process to take advantage of these different types of redundancy, considering intra-signal and inter-signal redundancies separately. We illustrate the process on vehicle telemetry signal data collected in a driver distraction monitoring project. We evaluate it using several machine learning algorithms: Random Forest; Naive Bayes; and C4.5 Decision Tree. Our results show that this two stage process significantly reduces the computation required because of inter-dependency calculations, while having minimal detrimental effect on the performance of the feature sets produced.

Reducing the data space and then classifying anomalies based on the reduced feature space is vital to real-time intrusion detection. In this study, a novel framework is developed for logistic regression-based anomaly detection and hierarchical feature reduction (HFR) to preprocess network traffic data before detection model training. The proposed dimensionality reduction algorithm optimally excludes the redundancy of features by considering the similarity of feature responses through a clustering analysis based on the feature space reduced by factor analysis, thus helping to rank the importance of input features (essential, secondary and insignificant) with low time complexity. Classification of anomalies over the reduced feature space is based on a multinomial logistic regression (MLR) model to detect multi-category attacks as an outcome with the goal of reinforcing detection efficiency. The proposed system not only achieves a significant detection performance, but also enables fast detection of multi-category attacks.

With the development of automatic management for industrial manufactory, the applications involving computer-vision and pattern recognition are widely used. The advantages of these modern methods using database to store a large sequence of images for helping management have been widely recognized. However, the storage of massive images into a database may demand a large memory space and cause a slow access speed. To increase the utilization rate of storage space and improve the performance of the database, this paper proposes an image compression scheme which is learnt from video compression to remove temporal and spatial redundancy in the image sequence. The proposed scheme not only alleviates the above issues associated with image storage but also keeps the basic database operations valid to image access, such as, insert, delete and update. At the end of this paper, we use the result to solve the optimizing storage problem successfully for a packaging machine which is computer-vision based quality monitoring.

In Brain-Machine Interfaces, the redudancy among neuron activities can decrease the generalization of a decoding model and increase computation complexity. This study proposes a novel algorithm based on symmetrical uncertainty that optimally selects neurons to decode kinetic variables. In this algorithm, both relevance and redundancy are considered. We tested the method by decoding spike train data recorded from a monkey using support vector regression. The results of extensive experiments show that only 15% of the recorded neurons can achieve a decoding performance comparable to using all neurons. The generalization of the selected neurons is no worse than using all neurons either. In addition, we have also compared our method with the traditional mutual information method and shown that the neurons selected by our method are less redundant.

Discriminative pattern mining is known under the names of subgroup discovery, contrast set mining, emerging pattern mining, etc. and has been intensively studied for the last 15 years. Based on the sophisticated techniques developed so far e.g. branch-and-bound search, minimum support raising, and redundancy elimination including the use of closed patterns, this paper proposes an efficient exact algorithm for finding top-k discriminative patterns that are not redundant and would be of value at a later step in prediction or knowledge discovery. The proposed algorithm is unique in that it conducts depth-first search over enumeration trees in a mirrored form of conventional ones, and by this design we can keep compact the list of candidate top-k patterns during the search and consequently high the minimum support threshold. Experimental results with the datasets from UCI Machine Learning Repository clearly show the efficiency of the proposed algorithm.

The wireless multimedia sensor network (WMSN) is an example of a platform for a multiple camera application. Since the devices in the WMSN are powered by batteries, energy consumption remains a critical issue. The previous chapters have discussed methods for single-view information reduction by removing the redundancy present in an image using event compression or by removing unimportant image frames using event detection. For multiple camera applications, additional redundancy can be found in cases where there is an overlap in the field of view (FOV) of different nodes. This is considered as the inter-camera redundancy that can be exploited to further reduce the data requirements for transmission. In this chapter, several approaches that have been proposed for multiple-view information reduction will be reviewed. The chapter begins by discussing approaches to remove or reduce the inter-camera redundancy. This is followed by a discussion on feature extraction which is a core technique used in searching for the similarities between two images. Then, an overview of the image stitching process will be provided. The performance of using different feature extraction methods to locate the same features in small size images and perform the stitching will be presented as well. The chapter concludes with an implementation of the proposed framework on the WMSN hardware platform.

Data Cleaning is a process for determining whether two or more records defined differently in database, represent the same real world object. Data Cleaning is a vital function in data warehouse preprocessing. It is found that the problem of duplication /redundancy is encountered frequently when large amounts of data collected from different sources is put in the warehouse. Eliminating redundancy in the data warehouse resolves conflicts in making wrong decisions. Data cleaning is also used to solve problem of “wastage of storage space”. One way of eliminating redundancy is by retrieving similar records using tokens formed on prominent attributes. Another approach is to use Conditional Functional Dependencies (CFD’s) to capture the consistency of data by combining semantically related data. Existing work on data cleaning do not deal with the case of multi-valued attributes. This paper deals with nesting based weak multi-valued dependencies (n-wMVD) which can handle multi-valued attributes and redundancy removal. Our contributions are of two fold (i) An approach to convert the given database to wMVD (ii) Implementation of n-wMVD to eliminate redundancy. The applicability of our approach was tested. The results are encouraging and are presented in the paper.

Critical systems require strict guarantees on information flow security and fault tolerance. We present a novel Boolean circuit methodology that can both monitor information flow throughout the hardware and simultaneously act as a triple modular redundant circuit. This is done by modifying the encoding technique for gate level information flow tracking (GLIFT). This new encoding not only has the added benefit of allowing for redundancy, it also reduces the size of the logic required for information flow tracking compared to the previous GLIFT encodings. This enables the development of high assurance systems on top of hardware with provable integrity and confidentiality properties. The new encoding also allows for these systems to be created with smaller area, lower power, and faster design time.

Nowadays the growing availability of stereo cameras for common applications is becoming a commodity. This paper addresses the problem of stereoscopic images data compression proposing an innovative algorithm for compressing Multi Picture Object coded stereopairs. By means of self organizing reconstruction algorithm based on image redundancy we are able to reduce the size of the enclosed JPEG images. The overall perceived (and measured) quality is managed by considering that a stereoscopic image represents the same scene acquired from two different perspectives. In particular we achieve some compression gain just encoding the two images with different quality factors. The reported results and test benchmarks show the robustness and efficiency of the proposed algorithm.

Growing user expectations of anywhere, anytime access to information require new types of data representations to be considered. While semi-structured data is a common exchange format, its verbose nature makes files of this type too large to be transferred quickly, especially where only a small part of that data is required by the user. There is consequently a need to develop new models of data storage to support the sharing of small segments of semi-structured data since existing XML compressors require the transfer of the entire compressed structure as a single unit. This paper examines the potential for bisimilarity-based partitioning (i.e. the grouping of items with similar structural patterns) to be combined with dictionary compression methods to produce a data storage model that remains directly accessible for query processing whilst facilitating the sharing of individual data segments. Study of the effects of differing types of bisimilarity upon the storage of data values identified the use of both forwards and backwards bisimilarity as the most promising basis for a dictionary-compressed structure. A query strategy is detailed that takes advantage of the compressed structure to reduce the number of data segments that must be accessed (and therefore transferred) to answer a query. A method to remove redundancy within the data dictionaries is also described and shown to have a positive effect on memory usage.

The Partial Redundancy Elimination (PRE) is an important redundancy elimination transformation technique. Locality based Static Single Assignment Partial Redundancy Elimination method classifies expressions into three categories: Local, Confined and Other. In order to reduce optimization time, LSSAPRE removes unnecessary Φ’s during the Φ -Insertion phase itself, which is the first step of SSAPRE. The unnecessary Φ – assignments for the local and confined expressions can be easily identified and removed in Φ –Insertion step. Even though this LSSAPRE system has reasonably good performance, it is not considering the other category expression based data flow problems in SSA form. In this paper, tried to propose a new Locality based SSAPRE method to remove unnecessary Φ assignments for the other category expressions during Φ - Insertion step itself and thereby reducing the analysis time and optimization time required for SSAPRE algorithm. But it theoretically proved that only based on the concept of locality of expressions and the dominance/post dominance relationship, it is not possible to remove the unnecessary Φ assignments for other category expressions during Φ - Insertion step and it is not possible to reduce the analysis time required for the SSAPRE algorithm.

In exploratory data mining it is important to assess the significance of results. Given that analysts have only limited time, it is important that we can measure this with regard to what we already know. That is, we want to be able to measure whether a result is interesting from a subjective point of view. With this as our goal, we formalise how to probabilistically model real-valued data by the Maximum Entropy principle, where we allow statistics on arbitrary sets of cells as background knowledge. As statistics, we consider means and variances, as well as histograms. The resulting models allow us to assess the likelihood of values, and can be used to verify the significance of possibly overlapping structures discovered in the data. As we can feed those structures back in, our model enables iterative identification of subjectively interesting structures. To show the flexibility of our model, we propose a subjective informativeness measure for tiles, i.e. rectangular sub-matrices, in real-valued data. The Information Ratio quantifies how strongly the knowledge of a structure reduces our uncertainty about the data with the amount of effort it would cost to consider it. Empirical evaluation shows that iterative scoring effectively reduces redundancy in ranking candidate tiles--showing the applicability of our model for a range of data mining fields aimed at discovering structure in real-valued data.

We develop a resolution-based method for forgetting concept and role symbols in \(\mathcal{ALCH}\) ontologies, or for computing uniform interpolants in \(\mathcal{ALCH}\). Uniform interpolants use only a restricted set of symbols, while preserving logical consequences of the original ontology involving these symbols. While recent work towards practical methods for uniform interpolation in expressive description logics limits attention to forgetting concept symbols, we believe most applications would benefit from the possibility to forget both role and concept symbols. We focus on the description logic \(\mathcal{ALCH}\), which allows for the formalisation of role hierarchies. Our approach is based on a recently developed resolution-based calculus for forgetting concept symbols in \(\mathcal{ALC}\) ontologies, which we extend by redundancy elimination techniques to make it practical for larger ontologies. Experiments on \(\mathcal{ALCH}\) fragments of real life ontologies suggest that our method is applicable in a lot of real-life applications.

Feature selection is an important data pre-processing step in pattern recognition and data mining. It is effective in reducing dimensionality and redundancy among the selected features, and increasing the performance of learning algorithm and generating information-rich features subset. In this regard, the chapter reports on a rough set-based feature selection algorithm called maximum relevance-maximum significance (MRMS), and its applications on quantitative structure activity relationship (QSAR) and gene expression data. It selects a set of features from a high-dimensional data set by maximizing the relevance and significance of the selected features. A theoretical analysis is reported to justify the use of both relevance and significance criteria for selecting a reduced feature set with high predictive accuracy. The importance of rough set theory for computing both relevance and significance of the features is also established. The performance of the MRMS algorithm, along with a comparison with other related methods, is studied on three QSAR data sets using the R 2 statistic of support vector regression method, and on five cancer and two arthritis microarray data sets by using the predictive accuracy of the K-nearest neighbor rule and support vector machine.

In this paper, we propose a novel anomaly targets detection algorithm baesd on information processing method and KRX anomaly detector. It use fully nolinear feature and decrease bands redundancy for hyperspectral imagery. Firstly, the original hyperspectral imagery is clustered by a new clustering method, i.e. k-clustering of particle swarm optimization. Then, we extract a largest fourth-order cumulant value in every class, and constitute a optimal band subset. Finally, the KRX detector is used on the band subset to get anomaly detection results. The simulation results demonstrate that the proposed PSOC-KRX algorithm outperforms the other algorithm, it is higher precision and lower false alarm rate.

The concepts of relevance and redundancy are central to feature selection algorithms that do not use a learning algorithm for subset evaluation. Redundancy is in fact a special form of relevance where there is a correlation (linear or nonlinear) between the input features of a problem. Therefore, having a good heuristic for measuring relevance can also help detect redundancy. In this paper, we show that there is a lack of generality in the solutions found by heuristic measures. Through some counter-examples we show that regardless of the type of heuristic measure and search strategy, filter methods cannot optimise the performance of all learning algorithms. We show how different measures may have different notions of relevance between features and how this could lead to not detecting important features in certain problems. We then propose a hyper-heuristic method that generates an appropriate relevance measure for each problem. The new approach can alleviate problems related to missing relevant features.

We discuss the singularity of the Fisher information arising from statistical inference for continuous-time stochastic processes of practical interest, such as asset price dynamics in finance and individual animal movement in biology, under high frequency discrete sampling schemes. Singularity seems to be caused by the scale parameter and the selfsimilarity index, while there exists a different type of singularity resulting from some redundancy of parameters in the short time framework. We derive the speed of convergence of the Fisher information to singularity for some instances and show that the convergence to singularity may be delayed through a wise expansion of the total observation window.

The use of mixture models in statistical analysis is increasing for dataset with heterogeneity and/or redundancy in the data. They are likelihood based models, and maximum likelihood estimates of parameters are attained by the use of the expectation maximization (EM) algorithm. Multi-modality of the likelihood surface means that the EM algorithm is highly dependent on starting points and poorly chosen initial points for the optimization may lead to only a local maximum, not a global maximum. The aim of this paper is to introduce a hybrid method of Particle Swarm Optimization (PSO) as a global optimization approach and the EM algorithm as a local search to overcome this problem and then it will be compared with different methods of choosing starting points in the EM algorithm.

In this paper, an experimental study of different ordinal regression methods and measures is presented. The first objective is to gather the results of a considerably high number of methods, datasets and measures, since there are not many previous comparative studies of this kind in the literature. The second objective is to detect the redundancy between the evaluation measures used for ordinal regression. The results obtained present the maximum MAE (maximum of the mean absolute error of the difference between the true and the predicted ranks of the worst classified class) as a very interesting alternative for ordinal regression, being the less uncorrelated with respect to the rest of measures. Additionally, SVOREX and SVORIM are found to yield very good performance when the objective is to minimize this maximum MAE.

High-availability Seamless Redundancy (HSR) is a redundancy protocol that provides duplicated frames for separate physical paths with zero recovery time. HSR is suitable for time critical applications in Industrial Ethernet Networks. However, the existing HSR scheme makes the unnecessary traffic by redundant frame copies. This drawback will degrade the network performance and may cause network congestion or delays. Thus, we propose a packet transmission scheme with different periods based on HSR ring topology for reducing network traffic load. Since it uses different transmission period by IEC 61850 traffic types, it can be adaptive for smart grid automation network technology. The main idea of the proposed scheme is using a different packet transmission period on each port of the HSR node. On one port near to destination, the node sends packets with the default transmission period, which is the same with the existing HSR scheme. On the other port, the node sends packets with an extended transmission period according to the delay requirement of IEC 61850 traffic types. By using the extended transmission period, the proposed scheme can reduce the network traffic and retain HSR network's availability. Also, we evaluate the proposed scheme by using OPNET simulation tool. The performance evaluation results show that the proposed scheme is better than the existing HSR scheme.

A large amount of data is involved in the processing of discrete data. As there is a large quantity of redundancy discrete data in mass data, it causes the error of discrete data association and reduced classification effect. To solve this problem, we brought forward the classification method of discrete data to improve associated mining algorithm. Before establishing association rule, we performed secondary validation of information which may contain redundancy data to lower down the probability of error in discrete data association, thus the possibility of false classification of data, and remove the defect of traditional method. As proved by simulation experiment, this improved algorithm is able to largely raise the preciseness of discrete data association and achieve better effect.

We propose a formal expansion of the transfer entropy to address the problem or partial conditioning evaluating information flow in multivariate datasets. This approach will then be adapted to put in evidence irreducible sets of variables which provide information for the future state of each assigned target. Multiplets characterized by an high value will be associated to informational circuits present in the system, with an informational character (synergetic or redundant) which can be associated to the sign of the contribution. These methods are then applied to the analysis of fMRI data.

To promote the efficiency of moving-object extraction using Gaussian Mixture Model, an improved method was proposed for intelligent video surveillance system. Based on the fact that the majority of the background pixels remain steady, these pixels were detected according to certain criteria before the procedures of matching and updating GMM parameters, and then the frequency of matching and updating was decreased. As a result, the computation redundancy could be reduced sharply. A couple of criteria to estimate the stable and unstable status as well as the flow chart were presented to illustrate the proposed method. Experimental results of two video sequences show that the speed of the improved method increase to 7.2 and 5.4 times, respectively, indicating that the efficiency is promoted significantly while keeping approximately equal quality.

We present an efficient algorithm for calculating $q$-gram frequencies on strings represented in compressed form, namely, as a straight line program (SLP). Given an SLP $\mathcal{T}$ of size $n$ that represents string $T$, the algorithm computes the occurrence frequencies of all $q$-grams in $T$, by reducing the problem to the weighted $q$-gram frequencies problem on a trie-like structure of size $m = |T|-\mathit{dup}(q,\mathcal{T})$, where $\mathit{dup}(q,\mathcal{T})$ is a quantity that represents the amount of redundancy that the SLP captures with respect to $q$-grams. The reduced problem can be solved in linear time. Since $m = O(qn)$, the running time of our algorithm is $O(\min\{|T|-\mathit{dup}(q,\mathcal{T}),qn\})$, improving our previous $O(qn)$ algorithm when $q = \Omega(|T|/n)$.

When using probabilistic neural network (PNN) to recognize human speaker, there exists structure complex problems if the training sample amount is large and the redundancy degree is high. To overcome this shortcoming, this paper proposes a method of principal component analysis (PCA) for keeping the effective information and reducing the redundancy of characteristic parameters, that means, this method can reduce the dimension of input data and optimize the structure of PNN network successfully. Experimental results show that the proposed speaker recognition method based on the combination of principal component analysis (PCA) and probabilistic neural network (PNN) is an effective and reliable new speaker recognition system.

We describe a technical solution for solving the following entailment problem in the Description Logic $\cal{EL}$: $({\cal T}\setminus U) \cup f^*({\cal T}\setminus U) \models C_1\sqsubseteq C_2$, where ${\cal T}$ is a TBox, C1,C2 are concepts, U is a subset of axioms of ${\cal T}$, and f* is a renaming function which gives a copy of TBox with some signature symbols injectively renamed. We introduce an enriched model structure for $\cal{EL}$ TBoxes: once built for a given TBox ${\cal T}$, it allows for checking the entailment efficiently for different choices of C1,C2, U, and f*. This problem is concerned with computing explicit definitions of concepts wrt a signature and is an important part of the recently proposed algorithms for ontology decomposition.

Subgroup discovery is a task at the intersection of predictive and descriptive induction, aiming at identifying subgroups that have the most unusual statistical (distributional) characteristics with respect to a property of interest. Although a great deal of work has been devoted to the topic, one remaining problem concerns the redundancy of subgroup descriptions, which often effectively convey very similar information. In this paper, we propose a quadratic programming based approach to reduce the amount of redundancy in the subgroup rules. Experimental results on 12 datasets show that the resulting subgroups are in fact less redundant compared to standard methods. In addition, our experiments show that the computational costs are significantly lower than the one of other methods compared in the paper.

We present an efficient algorithm for computing the LZ78 factorization of a text, where the text is represented as a straight line program (SLP), which is a context free grammar in the Chomsky normal form that generates a single string. Given an SLP of size $n$ representing a text $S$ of length $N$, our algorithm computes the LZ78 factorization of $T$ in $O(n\sqrt{N}+m\log N)$ time and $O(n\sqrt{N}+m)$ space, where $m$ is the number of resulting LZ78 factors. We also show how to improve the algorithm so that the $n\sqrt{N}$ term in the time and space complexities becomes either $nL$, where $L$ is the length of the longest LZ78 factor, or $(N - \alpha)$ where $\alpha \geq 0$ is a quantity which depends on the amount of redundancy that the SLP captures with respect to substrings of $S$ of a certain length. Since $m = O(N/\log_\sigma N)$ where $\sigma$ is the alphabet size, the latter is asymptotically at least as fast as a linear time algorithm which runs on the uncompressed string when $\sigma$ is constant, and can be more efficient when the text is compressible, i.e. when $m$ and $n$ are small.

We propose a new digital image watermarking method based on the dual-tree complex discrete wavelet transform (DT-CDWT) and interval arithmetic (IA). Both the DT-CDWT and IA produce redundancy from the original data. This implies that there is a possibility of developing a new watermarking method based on the DT-CDWT and IA. We describe our watermarking procedure in detail and show experimental results demonstrating that our method gives watermarked images that have better quality and that are robust against attacks such as marking, clipping, JPEG and JPEG2000 compressions, median filtering, addition of Gaussian white noise, addition of salt & pepper noise, rotation and resizing.

A variety of methods have been introduced to remove noise from digital images. However, many algorithms remove the fine details and structure of the image in addition to the noise because of assumptions made about the frequency content of the image. The non-local means algorithm does not make these assumptions, but instead assumes that the image contains an extensive amount of redundancy. This work will implement the non-local means algorithm and compare it to other denoising methods in experimental results. The main focus of this paper is to propose an improved non-local means algorithm addressing the preservation of structure in a digital image. The NL-means algorithm is proven to be asymptotically optimal under a generic statistical image model. The powerful evaluation method to be the visualization of the method noise on natural images.

In an image semantic annotation system, it often encounters the large-scale and high dimensional feature datasets problem, which leads to a slow learning process and degrading image semantic annotation accuracy. In order to reduce the high time complexity caused by redundancy information of image feature dataset, we adopt an improved affinity propagation (AP) algorithm to improve annotation by extracting and re-grouping the repeated feature points. The time consumption is reduced by square of repetition factor. The experiments results illustrate that the proposed annotation method has excellent time complexity and better annotation precision compared with original AP algorithms.

We develop an abstract model of information acquisition from redundant data. We assume a random sampling process from data which contain information with bias and are interested in the fraction of information we expect to learn as function of (i) the sampled fraction (recall) and (ii) varying bias of information (redundancy distributions). We develop two rules of thumb with varying robustness. We first show that, when information bias follows a Zipf distribution, the 80-20 rule or Pareto principle does surprisingly not hold, and we rather expect to learn less than 40% of the information when randomly sampling 20% of the overall data. We then analytically prove that for large data sets, randomized sampling from power-law distributions leads to "truncated distributions" with the same power-law exponent. This second rule is very robust and also holds for distributions that deviate substantially from a strict power law. We further give one particular family of powerlaw functions that remain completely invariant under sampling. Finally, we validate our model with two large Web data sets: link distributions to web domains and tag distributions on delicious.com.

We consider a modified version of the top-k subgroup discovery task, where subgroups dominated by other subgroups are discarded. The advantage of this modified task, known as relevant subgroup discovery, is that it avoids redundancy in the outcome. Although it has been applied in many applications, so far no efficient exact algorithm for this task has been proposed. Most existing solutions do not guarantee the exact solution (as a result of the use of non-admissible heuristics), while the only exact solution relies on the explicit storage of the whole search space, which results in prohibitively large memory requirements. In this paper, we present a new top-k relevant subgroup discovery algorithm which overcomes these shortcomings. Our solution is based on the fact that if an iterative deepening approach is applied, the relevance check - which is the root of the problems of all other approaches - can be realized based solely on the best k subgroups visited so far. The approach also allows for the integration of admissible pruning techniques like optimistic estimate pruning. The result is a fast, memory-efficient algorithm which clearly outperforms existing top-k relevant subgroup discovery approaches. Moreover, we analytically and empirically show that it is competitive with simpler approaches which do not consider the relevance criterion.

Window Memoization is a performance improvement technique for image processing algorithms. It is based on removing computational redundancy in an algorithm applied to a single image, which is inherited from data redundancy in the image. The technique employs a fuzzy reuse mechanism to eliminate unnecessary computations. This paper extends the window memoization technique such that in addition to exploiting the data redundancy in a single image, the data redundancy in a sequence of images of a volume data is also exploited. The detection of the additional data redundancy leads to higher speedups. The cascaded window memoization technique was applied to Canny edge detection algorithm where the volume data of prostate MR images were used. The typical speedup factor achieved by cascaded window memoization is 4.35x which is 0.93x higher than that of window memoization.

Finding relevant subspaces in very high-dimensional data is a challenging task not only for microarray data. The selection of features is to enhance the classification performance, but on the other hand the feature selection must be stable, i.e., the set of features selected should not change when using different subsets of a population. ensemble methods have succeeded in the increase of stability and classification accuracy. However, their runtime prevents them from scaling up to real-world applications.We propose two methods which enhance correlation-based feature selection such that the stability of feature selection comes with little or even no extra runtime.We show the efficiency of the algorithms analytically and empirically on a wide range of datasets.

Feature selection is a preprocessing step to machine learning, used to reduce the dimensionality of the dataset by removing irrelevant data. Variety of feature selection methods have been developed in the literature in order to increas the learning accuracy and reduce its complexity. In this paper we proposed a Bi-Layer behavioral-based feature selection approach. The proposed approach consists of two layers, in the first layer information gain is used to rank the features and select a new set of features depending on a global maxima classification accuracy. Then, in the second layer a new set of features is selected from within the first layer redacted data by searching for a group of local maximum classification accuracy in order to increase the number of reduced features. To evaluate the proposed approach it is applied on NSL-KDD dataset, where the number of features is reduced from 41 to 34 features in the first layer. Then reduced from 34 to 20 features in the second layer, which leads to improve the classification accuracy.

Due to the arbitrary motion patterns in practical video, annoying artifacts cased by the registration error often appears in the super resolution outcome. This paper proposes a spatial-temporal motion compensation based super resolution fusion method (STMC) for video after explicit motion estimation between a few neighboring frames. We first register the neighboring low resolution frames to proper positions in the high resolution frame, and then use the registered low resolution information as non-local redundancy to compensate the surrounding positions which have no or a few registered low resolution pixels. Experimental results indicate the proposed method can effectively reduce the artifacts cased by the motion estimation error with obvious performance improvement in both PSNR and visual effect.

Every data compression method assumes a certain model of the information source that produces the data. When we improve a data compression method, we are also improving the model of the source. This happens because, when the probability distribution of the assumed source model is closer to the true probability distribution of the source, a smaller relative entropy results and, therefore, fewer redundancy bits are required. This is why the importance of data compression goes beyond the usual goal of reducing the storage space or the transmission time of the information. In fact, in some situations, seeking better models is the main aim. In our view, this is the case for DNA sequence data. In this paper, we give hints on how finite-context (Markov) modeling may be used for DNA sequence analysis, through the construction of complexity profiles of the sequences. These profiles are able to unveil structures of the DNA, some of them with potential biological relevance.

Because of the development of wireless network, the error control and quality control under wireless network become an urgent problem to be resolved at present. In the paper, H.264 dynamic redundancy encoding algorithm based on the channel state and video correlation is proposed. By analyzing various mainstream redundant encoding algorithms, we adopted Pro MPEG FEC algorithms as the redundancy algorithm. In the algorithm that we proposed, redundancy bandwidth is allocated dynamically from various correlations of video content and channel condition and it can improve the utilization ratio of channel resources effectively. Thereby it can improve the quality of wireless channel video transmission. Simulation environment was built for the algorithms performance validation. The results of simulation experiments demonstrate that the algorithm can improve error recovery rate effectively under wireless network and the performance is better than the traditional channel redundancy algorithm, especially, when the network packet loss rate is changing.

Among many approaches to choosing the proper size of neural networks, one popular approach is to start with an oversized network and then prune it to a smaller size so as to attain better performance with less computational complexity. In this paper, a new hidden node pruning method is proposed based on the redundancy reduction among hidden nodes. The redundancy information is given by correlation coefficients among hidden nodes and this can save computational complexity. Experimental results demonstrate the effectiveness of the proposed method.

The classification of large number of object categories is a challenging trend in the Pattern Recognition field. In the literature, this is often addressed using an ensemble of classifiers . In this scope, the Error-Correcting Output Codes framework has demonstrated to be a powerful tool for combining classifiers. However, most of the state-of-the-art ECOC approaches use a linear or exponential number of classifiers, making the discrimination of a large number of classes unfeasible. In this paper, we explore and propose a compact design of ECOC in terms of the number of classifiers. Evolutionary computation is used for tuning the parameters of the classifiers and looking for the best compact ECOC code configuration. The results over several public UCI data sets and different multi-class Computer Vision problems show that the proposed methodology obtains comparable (even better) results than the state-of-the-art ECOC methodologies with far less number of dichotomizers.

Gene selection is to select the most informative genes from the whole gene set, which is a key step of the discriminant analysis of microarray data. Rough set theory is an efficient mathematical tool for further reducing redundancy. The main limitation of traditional rough set theory is the lack of effective methods for dealing with real-valued data. However, gene expression data sets are always continuous. This has been addressed by employing discretization methods, which may result in information loss. This paper investigates one approach combining feature ranking together with features selection based on tolerance rough set theory. Moreover, this paper explores the other method which can utilize the information contained within the boundary region to improve classification accuracy in gene expression data. Compared with gene selection algorithm based on rough set theory, the proposed methods are more effective for selecting high discriminative genes in cancer classification.

Intra prediction of H.264 encoder is to reduce spatial redundancy of image compression code. Intra prediction has up to 17 prediction modes, different modes have different ways to read reference pixel and to store. With so many prediction modes, how to deal with memory and how to read of reference pixels becomes more important. To reduce cycles of intra prediction, this paper employs an optimized hardware architecture, making intra 4×4 prediction and intra 16×16 prediction in parallel, and changes the order of intra 4×4 prediction. In this paper, a three-level storage structure of reference pixels is proposed, which saves storage resource and reduces access time. It takes at most 496 cycles to finish intra prediction of a MB in this paper and the execution time is reduced by 29.5%.

A video clip is composed of several moving images which have been recorded in time. In order to provide quality of service over the limited communication bandwidth, some compression is required. Compression is a redundancy reduction process where redundancies in visual data are identified and represented using much fewer binary digits or, hereafter bits. For instance, in a television program in which an anchor delivers news with her head and shoulders appeared on the static background, the only relevant and moving visual information are her head and shoulders, and the static background is unchanged for some time. Therefore, the moving head-and-shoulders and the static background can be coded with different compression strategies.

Feature selection is an important task in data analysis. mRMR is an equivalent form of the maximal statistical dependency criterion based on mutual information for first-order incremental supervised feature selection. This paper presents a novel feature selection criterion which can be considered as the unsupervised version of mRMR. The concepts of relevance and redundancy are both concerned in the feature selection criterion. The effectiveness of the new unsupervised feature selection criterion is confirmed by the theoretical proof. Experimental validation is also conducted on several popular data sets, and the results show that the new criterion can select features highly correlated with the latent class variable.

This paper focuses on borders and lossless representations for Constrained Datacubes of database relations, which can represent many-valued contexts. The final goal is to optimize both storage space and computation time. First we study the succinct representation through the borders Lower / Upper and Upper$^\sharp$ / Upper. However, these borders are not information-lossless. Therefore, by using the concept of cube closure, from a FCA perspective, we define three new information lossless representations for Constrained Datacubes: the L-Constrained Closed Datacube, the $U^\sharp$-Constrained Closed Datacube and the $U^{\sharp\sharp}$-Constrained Closed Cube. The latter is obtained by using the constrained closed tuples along with the border Upper$^\sharp$ from which redundancy is discarded in order to obtain an optimal representation. Finally, we evaluate experimentally the size of the introduced representations. The results are strongly emphasizing the idea that the latest structure is, to the best of our knowledge, the smallest representation of Constrained Datacubes.

For inter frame coding, the motion-compensated residual takes a large proportion of the total bits and the efficiency of the followed transform greatly affects the compression performance. In this paper, we propose an adaptive transform scheme to further exploit the non-local correlation for the motion-compensated residual. For a video sequence, there are usually repeating similar contents, especially between adjacent frames, known as temporal redundancy. We then use these content-similar blocks of the coding block, which most probably reflect the characteristic of the coding block, to train the adaptive transform. The predicted block together with the boundary reconstructed pixels of the coding block forms the target patch and is used to guide the searching of similar blocks. By fully exploring the correlation of abundant similar blocks, the proposed scheme achieves 0.1~0.5 dB gain in term of PSNR at high bit rate over the state-of-the-art scheme. For Mobile and BQSquare, 1dB gain is obtained at high bit rate.