Search

Network pruning has been widely adopted for reducing computational costs and memory consumption in low-resource devices. Recent studies show that the lottery ticket hypothesis achieves high accuracy under high compression ratios (i.e. , 80–90% of the parameters in original networks are removed). Nevertheless, finding well-trainable networks with sparse parameters (i.e. , < 10% of the parameters remaining) is still a challenging task, commonly believed by the lack of model capacity. This paper revisits the training process of existing pruning methods and observe that dead connections , which do not contribute to model capacity, occur in existing pruning methods. To this end, we propose a novel pruning method, namely all-alive pruning (AAP), which produces the pruned networks with only trainable weights with no dead connections. Notably, AAP is broadly applicable to various pruning methods and model architectures. We demonstrate that AAP equipped with the existing pruning methods (e.g. , iterative pruning, one-shot pruning, and dynamic pruning) consistently improves the accuracy of the original methods at high compression ratios on the various image- and language-based tasks. [ABSTRACT FROM AUTHOR]

Ensemble learning is a widely used methodology to build powerful predictors from multiple individual weaker ones. However, the vast majority of ensemble learning models are designed for offline application scenarios, the use of evolving fuzzy systems in ensemble learning for online learning from data streams has not been sufficiently explored, yet. In this paper, a novel self-adaptive fuzzy learning ensemble system is introduced for data stream prediction. The proposed ensemble system employs the very sparse random projection technique to compress the consequent parts of the learned fuzzy rules by individual base models to a more compressed form, thereby reducing redundant information and improving computational efficiency. To improve the overall prediction performance, a dynamical base model pruning scheme is introduced to the proposed ensemble system together with a novel inferencing scheme, such that less accurate base models will be removed from the ensemble structure at each learning cycle automatically and only these more accurate ones will be involved in joint decision-making. Numerical examples based on a wide range of benchmark datasets demonstrate the stronger prediction performance of the proposed ensemble system over the state-of-the-art alternatives. [ABSTRACT FROM AUTHOR]

Graph classification is crucial in network analyses. Networks face potential security threats, such as adversarial attacks. Some defense methods may trade off the algorithm complexity for robustness, such as adversarial training, whereas others may trade off clean example performance, such as smoothing-based defense. Most suffer from high complexity or low transferability. To address this problem, we proposed EGC2, an enhanced graph classification model with easy graph compression. EGC2 captures the relationship between the features of different nodes by constructing feature graphs and improving the aggregation of the node-level representations. To achieve lower-complexity defense applied to graph classification models, EGC2 utilizes a centrality-based edge-importance index to compress the graphs, filtering out trivial structures and adversarial perturbations in the input graphs, thus improving the model's robustness. Experiments on ten benchmark datasets demonstrate that the proposed feature read-out and graph compression mechanisms enhance the robustness of multiple basic models, resulting in a state-of-the-art performance in terms of accuracy and robustness against various adversarial attacks. [ABSTRACT FROM AUTHOR]

• New group theory based PSO learning for optimal parameters of fuzzy inference system. • Concept and an overview of the new algorithm for determining the filter coefficients. • Three-process analytical design for filter coefficient extraction to design person independent wavelet. • The energy compaction ratio is lower than the db4 coefficients. This paper presents a new method for designing matched digital filters with discrete valued coefficients. The fuzzy particle swarm optimization vector quantization (FPSOVQ) has been applied to obtain the optimum codebook in design of matched wavelet function. Abelian group has been used to extract the similarity present in the input voiced signal. Fuzzy particle swarm optimization (FPSO) process is used to find approximate ideal vector quantization (VQ) codebook to be carried out for compression of data. FPSOVQ scheme utilises features of fuzzy inference method (FIM) and expert particle swarm optimization (PSO). The generated codebook consists of set of highly ideal features, which are considered as the filter coefficients. These coefficients are used in the designing of the filter. All of the phonemes in the American English language were included in the 30 sentences that were chosen from the IEEE database. The sentences were originally down-sampled from 25 kHz to 8 kHz. The magnitude responses of each filter have been drawn, which indicates the characteristics of the filter. A comparison has been provided using energy compaction ratio as a parameter to judge the performance of matched designed filter with db4 filter. Experimental results show the advantage of the developed algorithm as the average value of energy compaction ratio for sampled voice signals is 2.8046 times lower for matched designed filter. [ABSTRACT FROM AUTHOR]

A text written using symbols from a given alphabet can be compressed using the Huffman code, which minimizes the length of the encoded text. It is necessary, however, to employ a text-specific codebook, i.e. the symbol-codeword dictionary, to decode the original text. Thus, the compression performance should be evaluated by the full code length, i.e. the length of the encoded text plus the length of the codebook. We studied several alphabets for compressing texts – letters, n -grams of letters, syllables, words, and phrases. If only sufficiently short texts are retained, an alphabet of letters or two-grams of letters is optimal. For the majority of Project Gutenberg texts, the best alphabet (the one that minimizes the full code length) is given by syllables or words, depending on the representation of the codebook. Letter 3 and 4-grams, having on average comparable length to syllables/words, perform noticeably worse than syllables or words. Word 2-grams also are never the best alphabet, on the account of having a very large codebook. We also show that the codebook representation is important – switching from a naive representation to a compact one significantly improves the matters for alphabets with large number of symbols, most notably the words. Thus, meaning-expressing elements of the language (syllables or words) provide the best compression alphabet. [ABSTRACT FROM AUTHOR]

Aiming at the deficiencies of existing image compression-encryption algorithms, such as unstable reconstruction performance and weak crypticity, a novel visually secure image encryption scheme based on the adaptive-thresholding sparsification compression sensing model and the newly designed memristive chaotic map is proposed in this paper. First, the plain image is sparsely represented in the wavelet packet domain. Then, the coefficient matrix is scrambled and compressed using parallel compressive sensing with assistance from the newly developed adaptive-thresholding sparsification model. Next, the compressed image is re-encrypted by the FAN transform and diffusion to generate a noise-like secret image. Finally, the reversible matrix encoding-based embedding method is employed to create a meaningful cipher image. To improve the execution efficiency of the encryption algorithm without affecting the security performance, a low-dimensional discrete memristive chaotic map is designed and applied in the compression, encryption and embedding stages. Additionally, the Lp matrix norm is utilized to establish an adaptive threshold model to effectively enhance the compression performance. Simulation results and analyses are presented to demonstrate the effectiveness, the compressibility, and the secrecy of the proposed scheme. [ABSTRACT FROM AUTHOR]

A novel joint image coding and reversible data hiding method for vector quantization (VQ) compressed images is proposed in this paper. Since the original VQ indices often exhibit uncorrelated, a rearrangement of them by considering their correlations may benefit the prediction performance so as to reduce the required bitrate. In this study, the tabu search algorithm is employed to rearrange codewords by fully exploiting their neighboring correlations, yielding a moepressible rearranged indices. By combining more highly-correlated indices of a to-be-predicted index into prediction, the improved linear regression method is then applied to achieve a sharper prediction-error histogram and less required additional information. After prediction, an adaptive run-length encoding method is presented to encode prediction errors, thereby eliminating unnecessary indicators. Experimental results demonstrate that the proposed method effectively reduces the bitrate of the compressed image while providing a comparable hiding capacity. [ABSTRACT FROM AUTHOR]

In large dynamic graphs, it is often impractical to store and process the entire graph. To contend with such graphs, lossless graph summarization is a compression technique that can provide a succinct graph representation without losing information. However, given a fully dynamic graph stream, the problem of lossless summarization is computationally and technically challenging. Although the state-of-the-art method performs well with respect to efficiency, its compression quality is usually unstable. This outcome occurs because it is a randomized algorithm and depends heavily on the pretuned parameters. In this paper, we propose a parameter-free approach to lossless summarization of a fully dynamic graph stream. In response to edge insertions and deletions, we first develop an incremental algorithm to maintain the summarization result by carefully exploring the influenced subgraph. Furthermore, we present a similarity-based algorithm to control the movement of vertices on the subgraph and thus guarantee the optimal result at each summarization update. To enhance the performance of our approach, we also propose two optimization techniques regarding candidate supernode refinement and destination supernode selection. The experimental results demonstrate that the proposed methods outperform the state-of-the-art methods by a large margin in terms of compression quality with comparable running time on the majority of datasets. [ABSTRACT FROM AUTHOR]

Filter pruning has been successfully applied in convolutional neural networks (CNN) to reduce model redundancy and speed up inference. Current pruning algorithms require setting a target value for a pruned model, which is not conducive to the rapid implementation in practical application. Firstly, for a target model, it may be necessary to pay attention to multiple performances at the same time, but it is hard to keep the best trade-off among indicators. Secondly, it is time-consuming to find a suitable model by pruning in a trial-and-error way. Therefore, this paper attempts to combine the most concerning indicators into a search target, rather than producing a model that only contains some pre-defined constraint. Specifically, the multi-object differential evolution algorithm is introduced into the search process, which is used to obtain a set of sparse structures that have taken all useful indicators into account. Then select proper sub-network to retrain from search results according to actual requirements. Experiments on multiple advanced CNN networks show that our algorithm can compress different indicators while keeping comparable or better performance than prior works. On ImageNet, pruning based on ResNet50 reduces FLOPs by 60.4%, while the accuracy of top-1 and top-5 only loses 1.09% and 0.57%, respectively. [ABSTRACT FROM AUTHOR]

Big Data is an emerging technology with enormous potential to develop business and its administration. Due to the enormous volume, efficient data mining and clustering methods are crucial to extracting meaningful insights and patterns from large-scale datasets. Problems may arise from the need to analyze, capture, share, store, and visualize the data. Several methods have already been proposed for mining knowledge from big data. It is practically inefficient or impossible to handle these massive data using the proposed methods in a single machine because big data are frequently acquired from dispersed locations and stored on several machines. Matrix decomposition is one of the critical strategies to retrieve knowledge from diverse, noisy, huge data generated by modern applications and stored in dispersed locations. This study proposes a novel approach called the Rank-Revealing QR Matrix and Schur Decomposition Method (RRQR-SDM) specifically designed for big data mining and clustering tasks. The RRQR-SDM is designed to reveal the rank of the data matrix in a computationally efficient manner by using a modified QR decomposition, eliminating the need for expensive Singular Value Decomposition (SVD) computations. The proposed RRQR-SDM method offers several advantages over existing approaches. Firstly, exploiting the inherent low-rank structure reduces the computational complexity associated with large-scale datasets. By revealing the rank of the input matrix, it enables dimensionality reduction and efficient data compression. Secondly, the Schur decomposition enhances the interpretability of the data by providing a clear separation between the relevant and irrelevant components. This feature makes the RRQR-SDM method particularly suitable for data mining and clustering tasks where identifying the most significant features is essential. To evaluate the performance of the RRQR-SDM method, extensive experiments were conducted on various big data datasets. The results demonstrate that the proposed method outperforms state-of-the-art computational efficiency and clustering accuracy techniques. [ABSTRACT FROM AUTHOR]

Prediction results in big data analysis can vary greatly depending on the data preprocessing methods used. Time series-based processing methods are particularly advantageous for prediction. While popular neural network models such as Back Propagation (BP), Recurrent Neural Network (RNN), and Long Short-Term Memory (LSTM) are based on weight, loss function, and other factors, their training efficiency is still relatively low. In this paper, we propose an efficient Time-series Recurrent Neural Network (TRNN) for stock price prediction. In the proposed model, trading volume is established and sliding windows are used to process the time series data. The trends and turning points of the data are extracted according to financial market features, and data compression is achieved. To improve the impact of recent trading volume on the current stock price, the price-volume relationship is upgraded from one dimension to two dimensions based on RNN. The information about trading volume is processed and compressed to establish the TRNN model, which guarantees both accuracy and efficiency. We compare our TRNN model with the original RNN and LSTM models in terms of efficiency and accuracy. We further discuss the feasibility of related expanded schemes of our TRNN model, as well as the extendability of the time-series compression and TRNN model to other fields. [ABSTRACT FROM AUTHOR]

Evolving fuzzy systems (EFSs) are a type of adaptive fuzzy rule-based systems which can self-adapt both their structures and parameters simultaneously. However, the existing EFSs suffer from two drawbacks: 1) classical EFSs usually use all input features to model systems, resulting in lengthy fuzzy rules; 2) some redundant information in fuzzy rules may hinder high generalization. To address these two issues, a promising method is proposed in this paper by combining very sparse random projection (VSRP) with a class of EFSs based-on data clouds, called VSRP-AnYa-EFS. The proposed method introduces: 1) a random sparse-Bernoulli (RSB) matrix based-on VSRP is utilized to compress the lengthy antecedent part into a tighter form, triggering a feature-reduction mechanism. By employing VSRP in RSB matrix, some redundant information in fuzzy rules can be filtered; 2) Local learning is used for consequent parameter optimization to suit decoupled behavior of rules after redundant information between rules is deleted. By adopting VSRP and local learning, the proposed VSRP-AnYa-EFS owns a compact structure and fast learning speed. Numerical examples presented in this paper demonstrate that the proposed method can significantly reduce training time from hours to minutes while the accuracy can be improved up to 5%. [ABSTRACT FROM AUTHOR]

The large-scale data stream problem refers to high-speed information flow which cannot be processed in scalable manner under a traditional computing platform. This problem also imposes expensive labelling cost making the deployment of fully supervised algorithms unfeasible. On the other hand, the problem of semi-supervised large-scale data streams is little explored in the literature because most works are designed in the traditional single-node computing environments while also being fully supervised approaches. This paper offers Weakly Supervised Scalable Teacher Forcing Network (WeScatterNet) to cope with the scarcity of labelled samples and the large-scale data streams simultaneously. WeScatterNet is crafted under distributed computing platform of Apache Spark with a data-free model fusion strategy for model compression after parallel computing stage. It features an open network structure to address the global and local drift problems while integrating a data augmentation, annotation and auto-correction (DA3) method for handling partially labelled data streams. The performance of WeScatterNet is numerically evaluated in the six large-scale data stream problems with only 25% label proportions. It shows highly competitive performance even if compared with fully supervised learners with 100% label proportions. [ABSTRACT FROM AUTHOR]

• We use several thresholds to convert different version of the data with varying certainty. • A random surfer model is employed to compute similarity values using higher order correlations. • Genetic algorithm is used as an optimization technique with the addition of a local search heuristic. • Our experiments show that using the proposed approach gives much better results even when uncertainly is quite high in the data. Probabilistic graphs have gained much interest in the data mining community since the big data revolution. Graph clustering is a widely used technique in exploratory data analysis such as data compression, information retrieval, protein–protein interaction network, etc. Traditional clustering algorithms assume that the data represented in a graph is deterministic, i.e., the edges between nodes are certain, though they may vary in their importance (weight). In probabilistic graphs, however, edges are uncertain and are represented as probabilities. In this work, we propose an evolutionary algorithm based clustering technique specifically to deal with uncertainties in such data. It has been observed that correlation usually exists among adjacent edges. Our work exploits this principle and explores the paths between nodes (including those of multiple order) to provide an estimate of the similarity between nodes with probabilistic edges. We embed this information into the fitness function and use Genetic algorithm to optimize the solution. The algorithm is tested on several benchmark datasets and compared with recent state-of-the-art algorithms. [ABSTRACT FROM AUTHOR]

Raster time series, a.k.a. temporal rasters, are collections of rasters covering the same region at consecutive timestamps. These data have been used in many different applications ranging from weather forecast systems to monitoring of forest degradation or soil contamination. Many different sensors are generating this type of data, which makes such analyses possible, but also challenges the technological capacity to store and retrieve the data. In this work, we propose a space-efficient representation of raster time series that is based on Compact Data Structures (CDS). Our method uses a strategy of snapshots and logs to represent the data, in which both components are represented using CDS. We study two variants of this strategy, one with regular sampling and another one based on a heuristic that determines at which timestamps should the snapshots be created to reduce the space redundancy. We perform a comprehensive experimental evaluation using real datasets. The results show that the proposed strategy is competitive in space with alternatives based on pure data compression, while providing much more efficient query times for different types of queries. [ABSTRACT FROM AUTHOR]

Currently, the demand for covert communication in open social networks brings new opportunities and challenges to existing image steganography technology in terms of robustness and security. To this end, an image robust adaptive steganography is proposed with robustness against multiple image processing attacks and detection resistance. First, a robust embedding domain with theoretical foundation and optimal invisibility is constructed based on the compression resistance principle. Then, utilizing the robust image abstraction and saliency measurement, the embedding channel is selected to avoid modifications in smooth regions and enhance visual quality. On this basis, the proposed method is given combining with error-correcting and STC codes to realize message embedding with minimum costs and improve extraction accuracy. Lastly, after parameters discussion and selection, the performance experiments are conducted compared with previous representative steganography algorithms, concerning robustness and detection resistance, and the fault tolerance is deduced, thereby providing the recommended coding parameters to improve message extraction integrity. The experimental results show that the proposed method can realize message extraction with high accuracy after JPEG compression, Gaussian noising, and scaling attacks, while holding comparable detection resistance to adaptive steganography against statistical features, which indicates its application prospect for covert communication in open lossy channels. [ABSTRACT FROM AUTHOR]

We are focused on improving the JPEG compression algorithm in the cases where a high ratio compression is required. For this purpose, we propose a preprocessing based on the F-transform. We develop a new discrete version of the inverse F-transform and new hybrid image compression and decompression algorithms, which combine JPEG with the F-transform. In the compression phase, JPEG follows the direct F-transform, while in the decompression phase, the methods are performed in the opposite order, and the new inverse F-transform is applied. We justify and illustrate on high-resolution benchmarks three main advantages of the proposed combination over the pure JPEG. We performed 504600 compressions using 15 different qualities, and for each selected quality it was confirmed that the proposed approach has compression 4 times stronger and the SSIM quality is 1.6 times higher than the pure JPEG. Finally, we propose an easy way to use the new compression software in the current JPEG standard. [ABSTRACT FROM AUTHOR]

The time series database is a specialized type of database specifically designed for storing and analyzing time series data. Compression of time series data is crucial for its performance. However, efficiently compressing time series data, particularly floating-point data, remains challenging. Existing compression algorithms are efficient for only a limited range of data patterns, indicating a lack of self-adaptation. In this paper, we propose an effective and A daptive lossless F loating-point C ompression algorithm AFC for time series databases. We devise four unique compression strategies, and based on the data patterns, AFC dynamically selects the appropriate strategy. These strategies handle data compression for diverse data patterns, enhancing the compression ratio and efficiency. The most suitable strategy is employed to achieve an optimal compression ratio. We compared our AFC algorithm with four state-of-the-art compression algorithms, namely Gorilla, FPC, TSXor, and Chimp, as well as various general-purpose compression algorithms such as LZ4 and Snappy. Experimental results demonstrate that our algorithm achieves an improvement of at least 20% in compression ratio and even up to 100% on certain datasets. [ABSTRACT FROM AUTHOR]

Nowadays, sensors and signal catchers in various fields are capturing time-series data all the time, and time-series data are exploding. Due to the large storage space requirements and redundancy, many compression techniques for time series have been proposed. However, the existing compression algorithms still face the challenge of the contradiction between random access and compression ratio. That is, in a time series database, large-scale time series data have high requirements on compression ratio, while large pieces of data need to be decompressed during the access process, resulting in poor query efficiency. In this paper, a proper solution is proposed to resolve such a contradiction. We propose a data compression method based on reinforcement learning, and use the idea of data deduplication to design the data compression method, so that the queries can be processed without decompression. We theoretically show that the proposed approach is effective and could ensure random accessing. To efficiently implement the reinforcement-learning-based solution, we develop a data compression method based on DQN network. Experiments show that the proposed algorithm performs well in time series data sets with large amount of data and strong regularity, performs well in compression ratio and compression time. Besides, since no decompression is required, the query processing time is much less than the competitors. [ABSTRACT FROM AUTHOR]

Multi-label classification aims to construct prediction models from input space to output space for multi-label datesets. However, the feature space of multi-label dataset and the hypothesis space of classifier are always complex. In order to obtain high performance multi-kernel learning algorithms, the core problem is to design algorithms that compress both hypothesis space and feature space. In this paper, by combining the local Rademacher complexity and Hilbert-Schmidt independence criterion (HSIC), an effective multi-kernel learning algorithm for multi-label classification is proposed to compress the feature space and hypothesis space simultaneously. Based on the tail sum of eigenvalues of the integral operator corresponding to kernels, the upper bound of local Rademacher complexity of linear functions class in the feature space is determined. The Hilbert-Schmidt independence criterion is applied to maximize the correlation between the convex combination of the input kernel matrices and the ideal kernel matrix in the label space. Finally, the Laplace regularization model is built by controlling the tail sum of eigenvalues of the integral operator corresponding to kernels of this criterion. Therefore, the above process obtains a combined kernel, simultaneously by sorting the coefficients of the combined kernel, a feature selection algorithm can be constructed. On the basis, the combined kernel can also be used to design a new binary classifier for multi-label classification. The experimental results verify that our proposed multi-kernel learning algorithms can effectively compress the hypothesis space, while the feature selection algorithms can effectively compress the feature space. [ABSTRACT FROM AUTHOR]

The flexible paradigm of Resource Description Framework (RDF) has accelerated the raw data published on the web. Therefore, the volume of generated RDF data has increased impressively in the last decade promoting compression to manage and reduce the size of RDF datasets. Universal RDF compressors can be able to detect and remove redundancy at symbolic, syntactic, or semantic levels. However, these compressors rarely exploit the graph patterns as well as structural regularities in real-world datasets. An efficient approach for compressing the RDF datasets based on the structural properties is HDT (Header-Dictionary-Triple). However, it cannot manage the RDF datasets with named graphs, the regularities of the graph structure, and structural redundancies. Because HDT considers all the triples to reside in the same default graph. Though, the triples of an RDF dataset belong to various named graphs. In this study, we have proposed a novel approach to deal with the above-mentioned challenges. We introduce hybrid TI-GI (Triple Interpreter-Graph Interpreter) to manage the RDF datasets with named graphs and use compact RDF serialization. We also propose RDF-RR (RDF-Redundancy Reducer) and object mapping that detects and removes structural redundancies by identifying the common patterns related to the predicates and objects in the RDF datasets. We employ a differential compressor to discover the frequent graph pattern in a single pass by using the data structure-oriented approach of the dataset. Evaluation of real-world datasets affirms that our proposed approach can substantially reduce the size of the experimental RDF datasets at approximately 30.52%, 24.92%, and 26.96% when compared with the existing HDT, HDT-FoQ (HDT-Focused on Querying) and the 2Tp (two Triple Predicate based index) approaches. Moreover, the indexing time of our proposed system is also reduced at approximately 17.89%, 13.70%, and 9.32% when compared with the HDT, HDT-FoQ, and 2Tp approaches. [ABSTRACT FROM AUTHOR]

• We propose an algorithm that extracts opcode execution sequences from a binary file without execution. • We formulate a CRNN, which detects malware using N extracted opcode sequences. • We provide an interpretation of the proposed model by analyzing the hidden layer activation of the CRNN. This paper presents a novel malware-detection model with a convolutional recurrent neural network using opcode sequences. Statistically, an executable file is considered as a set of consecutive machine codes. First, the theoretical foundation on which opcode sequences can be used to detect malware has been discussed. Next, an algorithm for extracting opcode sequences from executables and a deep learning-based malware-detection method that uses the opcode sequences as input have been presented. The proposed model comprises an opcode-level convolutional autoencoder that transforms a long opcode sequence to a relatively short compressed sequence at the front end and a dynamic recurrent neural network classifier that performs a prediction task using the codes generated by the opcode-level convolutional autoencoder at the rear end. Experimentally, the proposed model provided a malware-detection accuracy of 96 % , receiver operating characteristic-area under the curve of 0.99 , and true positive rate (TPR) of 95 %. The highest accuracy and TPR achieved by existing malware-detection methods using opcode sequences were 97 % and 82 % , respectively. Compared with this method, the proposed model delivered a slightly lower accuracy of 96 % but a considerably larger TPR of 95 %. Therefore, the proposed model is capable of more reliable malware detection. [ABSTRACT FROM AUTHOR]

Graph representations that preserve relevant topological information allow the use of a rich machine learning toolset for data-driven network analytics. Some notable graph representations in the literature are fruitful in their respective applications but they either lack interpretability or are unable to effectively encode a graph's structure at both local and global scale. In this work, we propose the Higher-Order Structure Descriptor (HOSD): an interpretable graph descriptor that captures information about the patterns in a graph at multiple scales. Scaling is achieved using a novel graph compression technique that reveals successive higher-order structures. The proposed descriptor is invariant to node permutations due to its graph-theoretic nature. We analyze the HOSD algorithm for time complexity and also prove the NP-completeness of three interesting graph compression problems. A faster version, HOSD-Lite, is also presented to approximate HOSD on dense graphs. We showcase the interpretability of our model by discussing structural patterns found within real-world datasets using HOSD. HOSD and HOSD-Lite are evaluated on benchmark datasets for applicability to classification problems; results demonstrate that a simple random forest setup based on our representations competes well with the current state-of-the-art graph embeddings. [ABSTRACT FROM AUTHOR]

Burrows-Wheeler Transform, followed by Move-To-Front Transform, are often used transformation techniques in data compression. They may reduce the information entropy of the input sequence, which becomes more compressible in this way. This paper suggests an alternative, a Prediction-based Move-To-Front Transform, which may replace the aforementioned transformations. According to the context, consisting of a few already seen symbols, the Prediction-based Move-To-Front Transform selects an appropriate ordered domain of symbols to achieve a better match with the currently transforming symbol. Freeman chain code in four and eight directions, Three-Orthogonal chain code, and Vertex Chain Code were used for experiments. We confirmed that the proposed approach, when using an appropriate length of context, reduces the information entropy to a similar extent as the Burrows-Wheeler Transform followed by the Move-To-Front Transform on chain code data. Both approaches led to a very similar compression efficiency on 32 testing shapes when an arithmetic coder was used in the final stage. The proposed approach turned out to be more efficient when longer chain code sequences were used, obtained by merging all the testing chain codes of the same type. [ABSTRACT FROM AUTHOR]

Reachability preserving compression generates small graphs that preserve the information only relevant to reachability queries, and the compressed graph can answer any reachability query without decompression. Existing reachability preserving compression algorithms either require a long compression time or include redundant data in the compressed graph. In this paper, we propose a novel edge sorting alogrithm for fast, non-redudant reachability preserving compression. On the other hand, we found that the current incremental reachability compression algorithms for dynamic graphs may return incorrect results in some cases. Therefore, we propose two novel incremental reachability compression algorithms. An algorithm called incremental reachability preserving compression with optimum compression ratio, which generates an update compressed graph that is exactly the same as the graph computed by recompression. The other algorithm called fast incremental reachability preserving compression, which can update the compressed graph in a short time. Extensive experiments on real datasets show the efficiency and the effectiveness of our methods. [ABSTRACT FROM AUTHOR]

This paper proposes a novel top-down approach to rule-based fuzzy systems, one that begins with the product—an equation—and then addresses the unique features of the product, without requiring the reader to know anything about fuzzy sets and systems. The "products" are adaptive variable-structure basis function expansions , where "adaptive variable-structure" means that different subsets of its basis functions are active (non-zero) in different regions of the state space, something that occurs automatically by virtue of the structure of the basis functions, so that the products can be said to "adapt" to locations in the state space. These products are novel candidates for machine learning. Unique features of all products are: (1) Number of basis functions is no longer a variable and is established locally through type-1 or type-2 uncertainty partitioning of each variable; (2) Both coarse and fine sculpting of the state space are achieved, and are described in terms of first- and second-order partitions of the state space, respectively; and (3) Linguistic interpretability is obtained, which may be of value to an end-user. Learning about rule-based fuzzy systems can be greatly compressed by using the top-down approach that is described in this paper. [ABSTRACT FROM AUTHOR]

In the era of big data, practical applications in various domains continually generate large-scale time-series data. Among them, some data show significant or potential periodicity characteristics, such as meteorological and financial data. It is critical to efficiently identify the potential periodic patterns from massive time-series data and provide accurate predictions. In this paper, a Periodicity-based Parallel Time Series Prediction (PPTSP) algorithm for large-scale time-series data is proposed and implemented in the Apache Spark cloud computing environment. To effectively handle the massive historical datasets, a Time Series Data Compression and Abstraction (TSDCA) algorithm is presented, which can reduce the data scale as well as accurately extracting the characteristics. Based on this, we propose a multi-layer time series periodic pattern recognition (MTSPPR) algorithm using the Fourier Spectrum Analysis (FSA) method. In addition, a Periodicity-based Time Series Prediction (PTSP) algorithm is proposed. Data in the subsequent period are predicted based on all previous period models, in which a time attenuation factor is introduced to control the impact of different periods on the prediction results. Moreover, to improve the performance of the proposed algorithms, we propose a parallel solution on the Apache Spark platform, using the Streaming real-time computing module. To efficiently process the large-scale time-series datasets in distributed computing environments, Distributed Streams (DStreams) and Resilient Distributed Datasets (RDDs) are used to store and calculate these datasets. Logical and data dependencies of RDDs in the P-TSDCA, P-MTSPPR, and P-PTSP processes are considered, and the corresponding parallel execution solutions are conducted. Extensive experimental results show that our PPTSP algorithm has significant advantages compared with other algorithms in terms of prediction accuracy and performance. [ABSTRACT FROM AUTHOR]

Highlights • A method for representing trajectories of moving objects in compressed space is presented. • The method is based on grammar-compressing the set of trajectories and associating summary information to the nonterminals of the grammar. • This representation supports various queries on the trajectories without decompressing the data. • It requires orders of magnitude less space than classical representations and is competitive in query times, outperforming classical alternatives in some cases. Abstract We introduce a compressed data structure for the storage of free trajectories of moving objects that efficiently supports various spatio-temporal queries. Our structure, dubbed GraCT, stores the absolute positions of all the objects at regular time intervals (snapshots) using a k 2-tree, which is a space- and time-efficient region quadtree. Positions between snapshots are represented as logs of relative movements and compressed using a grammar-based compressor. The non-terminals of this grammar are enhanced with MBR information to enable fast queries. The GraCT structure of a dataset occupies less than the raw data compressed with a powerful traditional compressor. Further, instead of requiring full decompression to access the data like a traditional compressor, GraCT supports direct access to object trajectories or to their position at specific time instants, as well as spatial range and nearest-neighbor queries on time instants and/or time intervals. Compared to traditional methods for storing and indexing spatio-temporal data, GraCT requires two orders of magnitude less space, and is competitive in query times. In particular, thanks to its compressed representation, the GraCT structure may reside in main memory in situations where any classical uncompressed index must resort to disk, thereby being one or two orders of magnitude faster. [ABSTRACT FROM AUTHOR]

This paper considers the use of interpolative coding for lossless chain code compression. The most popular chain codes are used, including Freeman chain code in eight (F8) and four directions (F4), Vertex Chain Code (VCC), and three-orthogonal chain code (3OT). The whole compression pipeline consists of the Burrows–Wheeler transform, Move-To-Front transform and the interpolative coding, which was improved by FELICS and new Ψ-coding. The approach was compared with the state-of-the-art chain code compression algorithms. For VCC, 3OT and F4, the obtained results are slightly better than the existing approaches. However, an important improvement was achieved with F8 chain code, where the presented approach is considerably better. [ABSTRACT FROM AUTHOR]

We present a novel subspace clustering algorithm SuMC (Subspace Memory Clustering) based on information theory, MDLP (Minimal Description Length Principle) and lossy compression. SuMC simultaneously solves two fundamental problems of subspace clustering: determination of the number of clusters and their optimal dimensions. Although SuMC requires only two parameters: data compression ratio r and a number of bits that are used to code a single scalar, the optimal value of compression ratio can be estimated by the Bayesian information criterion (BIC). We verified that in typical tasks of clustering, image segmentation and data compression, we obtain either better or comparable results to the leading methods of subspace clustering. [ABSTRACT FROM AUTHOR]

The human brain’s ability to efficiently detect patterns from the continuous streaming sensory stimuli has been a source of constant intrigue for naturalists, and has set the course for the development of research into artificial intelligence. Efficient encoding of such input stimuli into discrete timing of events play a decisive role in the ability of the spiking neurons inside the human brain to compress, transmit and recognise information presented by the external environment. In this article, we introduce the spike-time or temporal encoding paradigm as an efficient general approach to data compression for the purpose of pattern recognition. The data compression through spike-time encoding not only dramatically reduces the volume of data required to capture discriminatory information leading to economical storage and transmission, but can also be used for pattern recognition in streaming data domain. We experimentally show that the spike-time data produced by the temporal encoding techniques achieve comparable (superior in some cases) performance of pattern recognition in comparison to the use of the whole raw data. This article also introduces a generalised background knowledge driven optimisation based temporal encoding framework for encoding time series data and as an illustration of this approach, further formulates a temporal encoding algorithm, namely GAGamma, designed to efficiently compress fMRI data using discrete spike-times. We have evaluated the temporal encoding algorithms on the benchmark Starplus fMRI dataset, and the results demonstrate the temporal encoding algorithm’s ability to achieve significant data compression without sacrificing the performance of the pattern recognition in the compressed space. We also show that the GAGamma algorithm provides enhanced flexibility and control (compared to the state of the art temporal encoding algorithms) within the design framework of the temporal encoding problem for fMRI, leading to better quality of spike-time data. The proposed approach, by the use of efficient brain-like encoding mechanism, opens up new possibilities in information compression, communication and pattern recognition and thus is applicable to a range of new applications. [ABSTRACT FROM AUTHOR]

In this paper, we propose a lossless compression method to resolve the limitations in the real-time transmission of aurora spectral images. This method bi-dimensionally decorrelates the spatial and spectral domains and effectively removes side information of recursively computed coefficients to achieve high quality rapid compression. Experiments on data sets captured from the Antarctic Zhongshan Station show that the proposed algorithm can meet real-time requirements by using parallel processing to achieve outstanding compression ratio performance with low computational complexity. [ABSTRACT FROM AUTHOR]

The error-bounded Piecewise Linear Approximation (PLA) is to approximate the stream data by lines such that the approximation error at each point does not exceed a pre-defined error. In this paper, we focus on the version of PLA problem that generates connected lines in the segmentation for smooth approximation. We provide a new linear-time algorithm for the problem that outperform two of the existing methods with less number of connected segments. Our extensive experiments, on both real and synthetic data sets, indicate that our proposed algorithms are practically efficient. [ABSTRACT FROM AUTHOR]

Joining data compression and encryption is a way to keep secure data, as discussed by the current literature. While data compression responds to the great demand on data storage and transmission techniques, the encryption allows to handle some important parameters in a secure way. In wireless sensor networks the usual transform–based compression is the Discrete Wavelet Transform. In a previous paper we showed the good perfomance of the fuzzy transform (or F–trasform for short) based compression with respect to it. In this work, we propose a cubic B–spline F–transform in order to have a higher accuracy, even when data are not correlated, and a lower computational cost. Besides, in order to show the efficiency of the proposed approach, we compare it with the most recent lossless compression scheme in the field. We discuss these issues formally and numerically by using publicly available real–world data sets. The parameters required to decompress data are encrypted by means of a suitable existing encryption algorithm. We show that even if an illegal user had access to one of these parameters, our scheme would be still secure. [ABSTRACT FROM AUTHOR]

Environmental monitoring is one of the most important applications of wireless sensor networks (WSNs), which usually requires a lifetime of several months, or even years. However, the inherent restriction of energy carried within the battery of sensor nodes brings an extreme difficulty to obtain a satisfactory network lifetime, which becomes a bottleneck in scale of such applications in WSNs. In this paper, we propose a novel framework with dedicated combination of data prediction, compression, and recovery to simultaneously achieve accuracy and efficiency of the data processing in clustered WSNs. The main aim of the framework is to reduce the communication cost while guaranteeing the data processing and data prediction accuracy. In this framework, data prediction is achieved by implementing the Least Mean Square (LMS) dual prediction algorithm with optimal step size by minimizing the mean-square derivation (MSD), in a way that the cluster heads (CHs) can obtain a good approximation of the real data from the sensor nodes. On this basis, a centralized Principal Component Analysis (PCA) technique is utilized to perform the compression and recovery for the predicted data on the CHs and the sink, separately in order to save the communication cost and to eliminate the spatial redundancy of the sensed data about environment. All errors generated in these processes are finally evaluated theoretically, which come out to be controllable. Based on the theoretical analysis, we design a number of algorithms for implementation. Simulation results by using the real world data demonstrate that our framework provides a cost-effective solution to such as environmental monitoring applications in cluster based WSNs. [ABSTRACT FROM AUTHOR]

Reversible data hiding recovers the original images without distortion after the secret data are extracted, so it has received considerable attention among researchers in recent years. This work presents a novel reversible data hiding scheme based on the search-order coding (SOC) algorithm and side match vector quantization (SMVQ). The compression performance is improved by encoding the index with either SOC or SMVQ during the data embedding phase. The index of an image block also is encoded by using a few bits, thus improving the compression rate. During the extracting phase, the receiver uses simple steps to extract the secret data and reconstruct the original SMVQ compression code. Our experimental results indicated that, using a size 64 state codebook, the average compression rate with our scheme was 0.41 bpp, which is much better than the compression rates of Chang et al.’s, Yang and Lin’s, and Yang et al.’s schemes, i.e., 0.50, 0.54, and 0.56 bpp, respectively. Furthermore, in addition to reducing the time required for the embedding and extracting phases, the proposed scheme yields a higher embedding rate than the schemes of Yang and Lin and Yang et al. [ABSTRACT FROM AUTHOR]

In this paper, first a new optimum nonnegative integer bit allocation (ONIBA) algorithm is proposed for ‘transform coding’ with lower computational complexity compared to the existing algorithms. The algorithm is extended for the cases in which the variances of some sources are zero. Then, an ONIBA algorithm is developed for wavelet based ‘subband coding’. Finally the algorithm is modified for adaptive subband coding. Computer simulations demonstrate considerable superiority of the proposed subband coding algorithm over the existing ones. [ABSTRACT FROM AUTHOR]

Abstract: The frequent pattern tree (FP-tree) is an efficient data structure for association-rule mining without generation of candidate itemsets. It is used to compress a database into a tree structure which stores only large items. When data are modified, it, however, needs to process all transactions in a batch way. In the past, the prelarge-tree structure was proposed to incrementally mine association rules efficiently. In this paper, we propose an algorithm to maintain this structure when records in an original database are modified. The proposed maintenance algorithm is based on the pre-large concepts, which are defined by a lower support threshold and an upper support threshold. Due to the pruning properties of pre-large concepts, the proposed approach can reduce the rescan number of an original database when records are modified. It can thus obtain good execution performance for pre-large tree maintenance, especially when each time a small number of records are modified. Although experimental results show that the proposed prelarge-tree maintenance algorithm has good performance for handling modified records, the proposed algorithm needs to maintain nodes of pre-large items in the tree structure. This is the additional overhead, which is a trade-off between execution time and tree complexity. [Copyright &y& Elsevier]

Abstract: Information hiding for images is an important research issue in digital multimedia. It provides a way to certify copyrights of image files, as well as a means of secret communication for people. In this paper, we propose a reversible image information hiding scheme based on vector quantization (VQ). The proposed scheme uses a new technique called Adjoining State-Codebook Mapping (ASCM) to map the content of an image block to an index in the corresponding state-codebooks. The ASCM technique uses two adjacent image blocks of an encoding image block to create two state-codebooks, which can represent the encoding image block more accurately. Therefore, it can reduce the size of the output code-stream and yet still provide large embedding capacity for secret data. Experimental results indicate that the proposed scheme has better compression bit rate and larger embedding capacity when we compare the proposed scheme with recent similar studies, proposed by Chang et al. in 2009 and Yang et al. in 2011 respectively. [Copyright &y& Elsevier]

Abstract: Image forgery is becoming more prevalent in our daily lives due to advances in computers and image-editing software. As forgers develop more sophisticated forgeries, researchers must keep up to design more advanced ways of detecting these forgeries. Copy-move forgery is one type of image forgery where one region of an image is copied to another region in an attempt to cover a potentially important feature. This paper presents an efficient expanding block algorithm for detecting copy-move forgery and identifying the duplicated regions in an image. Experimental results show that the new method is effective in identifying size and shape of the duplicated region. Furthermore, it allows copy-move forgeries to be detected, where the copied region has been manufactured slightly lighter or darker, under JPEG compression, or with the effect of Gaussian blurring, in an attempt to throw-off detection algorithms. [Copyright &y& Elsevier]

Abstract: Motivated by the wavelet compression techniques and their applications, we consider the following problem: Given an unsorted array of numerical values and a threshold, what is the minimum number of elements chosen from the array, such that the sum of these elements is not less than the threshold value. In this article, we first provide two linear time algorithms for the problem. We then demonstrate the efficacy of these algorithms through experiments. Lastly, as an application of this research, we indicate that the construction of wavelet synopses on a prescribed error bound (in L 2 metric) can be solved in linear time. [Copyright &y& Elsevier]

Abstract: Many devices generate large amounts of data that follow some sort of sequentiality, e.g., motion sensors, e-pens, eye trackers, etc. and often these data need to be compressed for classification, storage, and/or retrieval tasks. Traditional clustering algorithms can be used for this purpose, but unfortunately they do not cope with the sequential information implicitly embedded in such data. Thus, we revisit the well-known K-means algorithm and provide a general method to properly cluster sequentially-distributed data. We present Warped K-Means (WKM), a multi-purpose partitional clustering procedure that minimizes the sum of squared error criterion, while imposing a hard sequentiality constraint in the classification step. We illustrate the properties of WKM in three applications, one being the segmentation and classification of human activity. WKM outperformed five state-of-the-art clustering techniques to simplify data trajectories, achieving a recognition accuracy of near 97%, which is an improvement of around 66% over their peers. Moreover, such an improvement came with a reduction in the computational cost of more than one order of magnitude. [Copyright &y& Elsevier]

Abstract: Subgraph mining is the process of identifying concepts describing interesting and repetitive subgraphs within graph-based data. The exponential number of possible subgraphs makes the problem very challenging. Existing methods apply a single-objective subgraph search with the view that interesting subgraphs are those capable of not merely compressing the data, but also enhancing the interpretation of the data considerably. Usually the methods operate by posing simple constraints (or user-defined thresholds) such as returning all subgraphs whose frequency is above a specified threshold. Such search approach may often return either a large number of solutions in the case of a weakly defined objective or very few in the case of a very strictly defined objective. In this paper, we propose a framework based on multiobjective evolutionary programming to mine subgraphs by jointly maximizing two objectives, support and size of the extracted subgraphs. The proposed methodology is able to discover a nondominated set of interesting subgraphs subject to tradeoff between the two objectives, which otherwise would not be achieved by the single-objective search. Besides, it can use different specific multiobjective evolutionary programming methods. Experimental results obtained by three of the latter methods on synthetically generated as well as real-life graph-based datasets validate the utility of the proposed methodology when benchmarked against classical single-objective methods and their previous, nonevolutionary multiobjective extensions. [Copyright &y& Elsevier]

Abstract: MP3, one of the most widely used digital audio formats, provides a high compression ratio with faithful quality. The widespread use enables MP3 audio files to become excellent covers to carry hidden information in audio steganography on the Internet. Our research, however, indicates that there are few steganalysis methods proposed to detect audio steganograms and that steganalysis methods for the information-hiding behavior in MP3 audio are particularly scarce. In this paper we propose a comprehensive approach to steganalysis of MP3 audio files by deriving a combination of features from quantized MDCT coefficients. We design frequency-based subband moment statistical features, accumulative Markov transition features, and accumulative neighboring joint density features on second-order derivatives. We also model the distortion by extracting the shape parameters of generalized Gaussian density from individual frames. Different feature selection algorithms are applied to improve detection accuracy. Signal complexity and modification density are introduced to provide a comprehensive evaluation. Experimental results show that our approach is successful in discriminating MP3 covers and the steganograms generated by the steganographic tool, MP3Stego, in each category of signal complexity, especially for the audio streams with high signal complexities that are generally more difficult to steganalyze. [Copyright &y& Elsevier]

Abstract: We address the problem of designing quantizers that cluster data while satisfying a k-anonymity requirement. A general data compression perspective is adopted, which considers both discrete and continuous probability distributions, and corresponding constraints on both cell sizes and quantizer index probabilities. Potential applications of this problem extend well beyond the important case of microdata anonymization, to include also optimized task allocation under workload constraints. Our contribution is twofold. First and most importantly, we present a theoretical analysis showing the optimality conditions which probability-constrained quantizers must satisfy, thereby theoretically characterizing optimal k-anonymous aggregation as a special case. As a second contribution, inspired by our theoretical analysis, we propose an alternating optimization algorithm for the design of this type of quantizers. Our algorithm is conceptually motivated by the popular Lloyd–Max algorithm for quantization design, originally intended for data compression, also known as the k-means method. Experimental results for synthetic and real data, with mean squared error as a distortion measure, confirm that our method outperforms MDAV, a popular fixed-size microaggregation algorithm for statistical disclosure control. This performance improvement is in terms of data utility, for the exact same k-anonymity constraint, but does come at the expense of higher computational sophistication. [Copyright &y& Elsevier]

Abstract: Data hiding is one of the protective techniques for authentication or secret communication through a public and open channel such as the Internet. In this study, a novel lossless data hiding by embedding secret data into a Vector Quantization (VQ)-compressed image is proposed to achieve secret communication and data compression simultaneously. The correlation of neighboring blocks of a VQ-compressed image is explored. It is shown that the neighboring blocks of a VQ-compressed image normally have high mutual correlation. Thus, this scheme employs the neighboring processed compression indices to generate specific sub-codebooks required for encoding and hiding data simultaneously. Since the sizes of sub-codebooks are smaller than that of the original VQ codebook, the encoded size of each index can be significantly reduced. As a result, a great deal of extra free space can be created. Moreover, the original VQ-compressed images can be perfectly recovered after data extraction. To evaluate the effectiveness of this approach, various test images are employed in the experiments. As documented in the experimental results, it is shown that the performance of the proposed scheme is superior to the former schemes in terms of compression ratio, embedding rate, execution time, and embedding capacity. [Copyright &y& Elsevier]

Abstract: Recently, a number of covering schemes proposed to exploit embedding capacity in covering images using matrix encoding. When binary secret data include a high frequency of repeated bit-blocks, a preprocessing phase is employed to compress secret data prior to the embedding process regardless of whether the compressed secret data still contain some repeated bit-blocks. This paper proposes two novel compact covering schemes, compact covering scheme (CCS) and intelligent compact covering scheme (ICCS), which directly compress and conceal the repeated bit-blocks of secret data bits into covering images within a single phase. The proposed schemes can exploit the embedding capacity and increase the embedding efficiency by reducing the required encoding operations. In addition, the proposed ICCS scheme can indicate up to which degree of repeated bit-blocks of secret data it can take advantage of in order to improve the embedding capacity. Experimental results demonstrate that the proposed ICCS method can significantly improve the embedding capacity when the ratio of the repeated bit-blocks over the total number of blocks is low. The proposed schemes are feasible for actual data-hiding applications. [Copyright &y& Elsevier]

Abstract: Previous work has shown that the Minkowski-p distance metrics are unsuitable for clustering very high dimensional document data. We extend this work. We frame statistical theory on the relationships between the Euclidean, cosine, and correlation distance metrics in terms of item neighborhoods. We discuss the differences between the cosine and correlation distance metrics and illustrate our discussion with an example from collaborative filtering. We introduce a family of normalized Minkowski metrics and test their use on both document data and synthetic data generated from the uniform distribution. We describe a range of criteria for testing neighborhood homogeneity relative to underlying latent classes. We discuss how these criteria are explicitly and implicitly linked to classification performance. By testing both normalized and non-normalized Minkowski-p metrics for multiple values of p, we separate out distance compression effects from normalization effects. For multi-class classification problems, we believe that distance compression on high dimensional data aids classification and data analysis. For document data, we find that the cosine (and normalized Euclidean), correlation, and proportioned city block metrics give strong neighborhood recovery. The proportioned city block metric gives particularly good results for nearest neighbors recovery and should be used when utilizing document data analysis techniques for which nearest neighborhood recovery is important. For data generated from the uniform distribution, neighborhood recovery improves as the value of p increases. [Copyright &y& Elsevier]

Abstract: Data compression has been widely applied in many data processing areas. Compression methods use variable-length codes with the shorter codes assigned to symbols or groups of symbols that appear in the data frequently. There exist many coding algorithms, e.g. Elias-delta codes, Fibonacci codes and other variable-length codes which are often applied to encoding of numbers. Although we often do not consider time consumption of decompression as well as compression algorithms, there are cases where the decompression time is a critical issue. For example, a real-time compression of data structures, applied in the case of the physical implementation of database management systems, follows this issue. In this case, pages of a data structure are decompressed during every reading from a secondary storage into the main memory or items of a page are decompressed during every access to the page. Obviously, efficiency of a decompression algorithm is extremely important. Since fast decoding algorithms were not known until recently, variable-length codes have not been used in the data processing area. In this article, we introduce fast decoding algorithms for Elias-delta, Fibonacci of order 2 as well as Fibonacci of order 3 codes. We provide a theoretical background making these fast algorithms possible. Moreover, we introduce a new code, called the Elias–Fibonacci code, with a lower compression ratio than the Fibonacci of order 3 code for lower numbers; however, this new code provides a faster decoding time than other tested codes. Codes of Elias–Fibonacci are shorter than other compared codes for numbers longer than 26 bits. All these algorithms are suitable in the case of data processing tasks with special emphasis on the decompression time. [Copyright &y& Elsevier]

Abstract: This paper presents Edge-Guided (E-G), an optimized text preprocessing technique for compression purposes. It transforms the original text into a word net, which stores all relationships between adjoining words. A specific directed graph is proposed to model this transformation: words are stored in vertices, whereas edges represent word transitions. Thus, the word net has a text representation which reflects the natural word order in the text, so it can be used directly for encoding purposes. A specific coding scheme is described at the top of the word net. It regards a text as a sequence of word transitions, in such a way that each word is encoded by traversing a specific edge from the vertex which stores the preceding word. This accomplishes a 1-order text preprocessing whose output is an intermediate byte representation that can be effectively encoded with universal techniques. This technique is called E-G 1 and performs on some variants. This experience is used to revisit the concept of word net. It is used to identify significative 2-word symbols by performing a specific transformation on frequent edges. The resultant transformed word net appends these 2-word symbols to the original word vocabulary, and allows a specific hierarchical relationship between them and their component words. The transformed approach also enhances the original coding scheme to handle these new features. A new technique, called E-G 2, approximates a 2-order model on words that also support specific variants. Both techniques are studied from empirical and experimental perspectives. Some compressors are also used to analyze the preprocessing ability of E-G with respect to different compression approaches. Competitive space/time trade-offs are achieved when our approaches are used to compress medium-large size texts. The best results are achieved when E-G preprocessing is coupled with high-order compressors such as Prediction by Partial Matching (PPM). [Copyright &y& Elsevier]