Lossless image compression algorithms perform cross-channel decorrelation transformations, which are called reversible color transforms (RCTs), usually using integer lifting steps, which either require bit width inflated by one bit for the chroma channels or the use modular arithmetic that introduces wraparound artifacts, resulting from integer overflows. In this work we propose a fundamental technique that solves these problems entirely, by fusing the RCT with spatial prediction and adding causality to color domain, by predicting image data in a permuted RGB space. This method entirely avoids overflows, maintains reversibility, produces residuals with the same word size as the input, supports any kind of scan-line spatial predictors, and can be generalized to any number of channels. The proposed transformation is even guaranteed to improve coding efficiency, compared to all known fixed RCTs, without increasing computational cost or memory usage.

Convolutional neural networks (CNNs) have been widely used with remarkable success in the acoustic scene classification (ASC) task. However, the performance of these CNNs highly relies on their architectures, requiring a lot of effort and expertise to design CNNs suitable for the investigated problem. In this work, we propose an efficient genetic algorithm (GA) that aims to find optimized CNN architectures for the ASC task. The proposed algorithm uses frequency-dimension splitting of the input spectrograms in order to explore the architecture search space in sub-CNN models in addition to classical single-path CNNs. Specifically, this algorithm aims to find the best number of sub-CNNs in addition to their architectures to better capture the distinct features of the input spectrograms. The proposed GA is specifically designed for sound classification to suit the ASC task than many other GAs that optimize conventional single-path CNN architectures. Experimental results on three benchmark datasets demonstrate the effectiveness of the proposed method. Specifically, the proposed algorithm has achieved around 17.8%, 16%, and 17.2%, relative improvement in accuracy with respect to the baseline systems on the development datasets of DCASE2018-Task1A, DCASE2019-Task1A, and DCASE2020-Task1A, respectively.

Abstract Live migration is an essential feature in virtual infrastructure and cloud computing datacenters. Using live migration, virtual machines can be online migrated from a physical machine to another with negligible service interruption. Load balance, power saving, dynamic resource allocation, and high availability algorithms in virtual data-centers and cloud computing environments are dependent on live migration. Live migration process has six phases that result in live migration cost. Several papers analyze and model live migration costs for different hypervisors, different kinds of workloads and different models of analysis. In addition, there are also many other papers that provide prediction techniques for live migration costs. It is a challenge for the reader to organize, classify, and compare live migration overhead research papers due to the broad focus of the papers in this domain. In this survey paper, we classify, analyze, and compare different papers that cover pre-copy live migration cost analysis and prediction from different angels to show the contributions and the drawbacks of each study. Papers classification helps the readers to get different studies details about a specific live migration cost parameter. The classification of the paper considers the papers’ research focus, methodology, the hypervisors, and the cost parameters. Papers analysis helps the readers to know which model can be used for which hypervisor and to know the techniques used for live migration cost analysis and prediction. Papers comparison shows the contributions, drawbacks, and the modeling differences by each paper in a table format that simplifies the comparison. Virtualized Data-center and cloud computing clusters admins can also make use of this paper to know which live migration cost prediction model can fit for their environments.

Driver distraction identification is crucial to improve road safety. Through vehicular communications, vehicles can exchange driver behavior information, based on which distracted drivers can be identified, and all drivers can be notified, which can mitigate the impact of driver distraction on road safety. To build such systems, it is essential to understand the different types of distraction, and how they affect driver behavior and their relationship to crashes or near-crashes. This understating should be based on real datasets, which are very limited. Therefore, in this paper, we cover this gap by building a data-driven simulation model to quantify the impact of realistic driver distraction on traffic safety. In particular, we use the 2nd Strategic Highway Research Program Naturalistic Driving Study (SHRP2 NDS) dataset to develop a simulation framework for driver distraction. First, we pre-process and analyze the dataset for different types of distractions. The analysis shows that the data can not be fitted to any of the known distributions. Therefore, we use the Gaussian Mixture Model (GMM) to represent the distraction intervals for the different distraction types. We then use these GMM models and the statistics collected from the data to realistically simulate the driver distraction using the Simulation for Urban MObility (SUMO) software. Finally, we use this framework to simulate the driver distraction in a real network. The data analysis and simulation results revealed important and interesting conclusions, such as decreasing the crash ratio when roads become congested.

The past decade has witnessed a staggering evolution in cellular networks. Mobile wireless technologies have undergone four distinct generations; from uncomplicated voice calls in the first generation to high-speed, low latency and video streaming in the fourth generation. The numerous services brought to the users by 4G network have caused an increasing load demand. This increasing demand in network usage has proven the necessity of further service enhancements, such as predictive resource allocation techniques and handover analysis. For these techniques to be deployed, network quality and performance analysis must be performed on real-world network data. Since throughput is a major indicator of the network's performance, throughput modelling and prediction can be utilized for analyzing network quality. In this paper, two approaches for throughput analysis are examined: classical machine learning and time series forecasting. For the first approach, various machine learning models were deployed for throughput prediction and our analysis showed that the random forest model achieved the highest prediction performance. For time series forecasting, statistical methods as well as deep learning architectures were used. The evaluation shows that the machine learning models had a higher throughput prediction performance than the time series forecasting techniques.

With the advancements in wireless network technologies over the past few decades and the deployment of 4G LTE networks, the capabilities and services provided to end-users have become seemingly endless. Users of smartphones utilize high-speed network services while commuting on public transit and hope to have a consistent, high-quality connection for the duration of their trip. Due to the massive load demand on cellular networks and frequent changes in the underlying radio channel, users often experience sudden unexpected variations in the connection quality. To overcome such variations and maintain a consistent connection, these variations need to be predicted before they occur. This can be accomplished by the spatio-temporal analysis of the different network quality parameters and the investigation of the main factors that affect the network's performance and QoS. To this end, we conducted a network survey via Kingston Transit in Kingston, Ontario, Canada. We used the Android network monitoring application G-NetTrack Pro to build a dataset of various client-side wireless network quality parameters. The dataset consists of 30 repeated public transit bus trips at three different times of the day, each lasting around one hour. In this paper, we describe the data collection process, present an analysis of the collected data, and investigate the effects of time and location on the network's measured throughput and signal strength. We made the collected data, including more than 190 thousand unique records, publicly available to researchers in a domain where open data is rare.

Virtualization became a commonly used technology in datacenters during the last decade. Live migration is an essential feature in most of the clusters hypervisors. Live migration process has a cost that includes the migration time, downtime, IP network overhead, CPU overhead and power consumption. This migration cost cannot be ignored, however datacenter admins do live migration without expectations about the resultant cost. Several research papers have discussed this problem, however they could not provide a practical model that can be easily implemented for cost prediction in VMware environments. In this paper, we propose a machine learning approach for live migration cost prediction in VMware environments. The proposed approach is implemented as a VMware PowerCLI script that can be easily implemented and run in any vCenter Server Cluster to do data collection of previous migrations statistics, train the machine learning models and then predict live migration cost. Testing results show how the proposed framework can predict live migration time, network throughput and power consumption cost with accurate results and for different kinds of workloads. This helps datacenters admins to have better planning for their VMware environments live migrations.

This paper introduces an approach for fast tracking, quantization and centreline extraction of small vascular structures in medical imaging, with the problem of coronary segmentation used as an example. The tracking is achieved by propagating an explicit surface. This surface is represented as a triangular adaptive mesh where the element size is adjusted based on the current size of the vessel. Adaptive re-meshing is performed on-the-fly during propagation in an efficient manner. An effective self-intersection prevention method is introduced to address one of the major issues in triangular mesh offsetting.

Key exchange is one of the major concerns in cryp-tology. Neural cryptography is a recent non-classical paradigm which achieves key exchange by mutual learning between two neural networks that receive the same input patterns and update their weights using specific rules. Each weight component of the network can be seen as a random walker in the weight space. The two walkers move in the weight space and reflect at two boundaries (left and right) which represent the network synaptic depth. The reflecting boundaries cause the distance between the two walkers decreases if one of them hits the boundary when a common direction is chosen at each step. Therefore, the mutual learning algorithm relies on this defined boundary condition to achieve synchronization between the two parties. In this paper, we aim to increase the security of the neural cryptography by authenticating the communication using preshared secrets. The mutual learning algorithm is modified so that the reflecting boundaries become hidden and only accessible by the two partners. New update rules are developed to exploit the secret information without adding any limitation to the initial configuration for the two parties. This is done by converting the two boundaries located at a straight line path to a one secret boundary located randomly at a circular path. Therefore, the mutual learning is impeded except this secret information is known. The proposed algorithm is called Neural Cryptography with Secret Boundaries (NCSB) and it is proved with information theory that the secret boundaries can not be revealed from the public information broadcast through the public channel.

This paper presents a mixed coding evolutionary algorithm for learning Gaussian mixture models. The proposed algorithm can find the optimal number of mixture components in addition to the various mixture parameters that include the mixing probabilities, mean vectors and covariance matrices. This is achieved by devising a mixed-coded genetic algorithm that encodes the mixture parameters into its chromosomes that will undergo different genetic operators that maximize a model-based fitness function. The likelihood of the observed data is maximized while the Akaike Information criterion (AIC) will be used to produce the minimum model structure.

Neural cryptography deals with the problem of "key exchange" between two neural networks using the mutual learning concept. The two networks exchange their outputs (in bits) and the key between the two communicating parties is eventually represented in the final learned weights, when the two networks are said to be synchronized. Security of neural synchronization is put at risk if an attacker is capable of synchronizing with any of the two parties during the training process. Therefore, diminishing the probability of such a threat improves the reliability of exchanging the output bits through a public channel. The synchronization with feedback algorithm is one of the existing algorithms that enhances the security of neural cryptography. This paper proposes three new algorithms to enhance the mutual learning process. They mainly depend on disrupting the attacker confidence in the exchanged outputs and input patterns during training. The first algorithm is called "Do not Trust My Partner" (DTMP), which relies on one party sending erroneous output bits, with the other party being capable of predicting and correcting this error. The second algorithm is called "Synchronization with Common Secret Feedback" (SCSFB), where inputs are kept partially secret and the attacker has to train its network on input patterns that are different from the training sets used by the communicating parties. The third algorithm is a hybrid technique combining the features of the DTMP and SCSFB. The proposed approaches are shown to outperform the synchronization with feedback algorithm in the time needed for the parties to synchronize.

This work outlines an evolutionary algorithm for image vector quantization. An integer-coded genetic algorithm (GA) that employs the maximum likelihood (ML) measure as the fitness function is introduced. The proposed algorithm allows for different chromosome representations and provides an adaptation to the genetic operators to suit the image quantization problem. The main objective of the algorithm is, for a codebook with a pre-defined size, to find the best set of image blocks that make up the codewords. Each codeword will be representative of a group of blocks.

This work describes an enhanced technique for on-line signature verification. The distance between two signatures is computed by dynamic time warping (DTW) method. The reference signatures are used to assign special parameters for each signer, which makes the system cover the intra signer variation. Several features are extracted. Systems with single and multi-features are tested. Curvature change and speed enhance success verification rate. The experiments have been carried out using the SUSIG online signature database. The best result for ROC area under curve is 99.5 with equal error rate 3.48%, and the best result for equal error rate is 3.06% with ROC area under curve 99.43.

Neural cryptography deals with the problem of key exchange using the mutual learning concept between two neural networks. The two networks will exchange their outputs (in bits) so that the key between the two communicating parties is eventually represented in the final learned weights and the two networks are said to be synchronized. Security of neural synchronization depends on the probability that an attacker can synchronize with any of the two parties during the training process, so decreasing this probability improves the reliability of exchanging their output bits through a public channel. This work proposes an exchange technique that will disrupt the attacker confidence in the exchanged outputs during training. The algorithm is based on one party sending erroneous output bits with the other party being capable of predicting and removing this error. The proposed approach is shown to outperform the synchronization with feedback algorithm in the time needed for the parties to synchronize.

In this paper, a functional link neural net (FLN) capable of performing sparse nonlinear system identification is proposed. Korenberg's fast orthogonal search (FOS) is adopted to detect the proper model and its associated parameters. The FOS algorithm is modified by first sorting all possible nonlinear functional expansion of the input pattern according to their correlation with the system output. The sorted functions are divided into equal size groups, pins, where functions with the highest correlation with the output are assigned to the first pin. Lower correlation members go the following pin and so forth. During the identification process, members in lower pins are tried first. If a solution is not found, next pins join the candidates pool until the identification process completes within prespecified accuracy. The modified Gram Schmidt orthogonalization and Choleskey decomposition are applied to create orthogonal functionals that can linearly fit the identified system. The proposed architecture is tested on noise-free and noisy nonlinear systems and shown to find sparse models that can approximate the experimented systems with acceptable accuracy.

Vector quantization (VQ) has been extensively used in speech vocoders. The training process normally requires a very large training-set. This paper introduces a novel particle-based distortion measure for the high-band speech spectrum that enables the vector quantizer to operate given a relatively small training-set. This measure has been used in the construction of a segmental vocoder, which is using the pitch period as segments. A description of the proposed particle approach, the energy-mass distortion measure, is given. A comparison between the energy-mass and the use of the mel-frequency cepstral coefficients (MFCC) as a distortion measure shows the ability of the particle approach to better represent the speech formants, when operating under the small training-set constraint. Finally, the performance of the energy-mass is evaluated using the spectral distortion (SD). Speech quality perceived by the receiver is evaluated using the recently standardized objective quality measure PESQ where an improvement of 0.3 PESQ score was obtained.

In this paper, an integer-coded genetic algorithm (GA) is proposed to obtain accurate and parsimonious sinusoidal representation of signals. The method suggested addresses the problem of finding the most significant sinusoidal frequencies which lead to the smallest modeling error. A variable length GA chromosome which encodes the number of possible component frequencies and their actual locations is used. The weight of each component is calculated using a least squares identification method. A set of evolutionary operators are selected to suit the integer representation of the genes in each chromosome. The proposed algorithm produced excellent results in modeling noise-free and noisy time series of short data records.

Presented in this paper is a tool that automatically migrates analog designs from one process to another while keeping circuit and layout topologies. A netlist migration engine recalculates the new device dimensions in the target technology followed by a layout migration engine that compacts the design according to the new process design rules. The overall framework preserves design intelligence embedded in the original IP such as symmetry, hierarchy, placement and routing. The circuit migration engine, being very fast, can retarget large analog blocks in only a few minutes while giving same or better performance of the original design. The migration of 3 different circuits is presented to validate the overall methodology. These circuits have been fabricated and measured.

This paper presents a field programmable gate array (FPGA) implementation for a competitive Hopfield neural network (CHNN) to be used in image histogram equalization (HE). This algorithm is so computationally expensive that a viable hardware implementation is appealing provided that an efficient algorithm-to-architecture mapping can be achieved. The Xilinx Virtex-E was used for the hardware realization. An efficient use of the chip is outlined in the results.

Routing FPGAs (Verma, 1999) is a challenging problem because of the relative scarcity of routing resources represented in wires and connection points. This can lead either to slow implementations caused by long wiring paths that avoid congestion or a failure to route all signals (McMurchie, 1995). This paper presents some enhancements to improve both the performance and routability of the routability-driven routing using the versatile place and route VPR routing tool (Betz, 1999). Testing MCNC benchmarks has shown the efficiency of these enhancements.

In this paper, a floating-point genetic algorithm (GA) for Volterra-system identification is presented. The adaptive GA method suggested here addresses the problem of determining the proper Volterra candidates, which leads to the smallest error between the identified nonlinear system and the Volterra model. This is achieved by using variable-length GA chromosomes, which encode the coefficients of the selected candidates. The algorithm relies on sorting all candidates according to their correlation with the output. A certain number of candidates with the highest correlation with the output are selected to undergo the first evolution "era". During the process of evolution the candidates with the least significant contribution in the error-reduction process is removed. Then, the next set of candidates are applied into the next era. The process continues until a solution is found. The proposed GA method handles the issues of detecting the proper Volterra candidates and calculating the associated coefficients as a nonseparable process. The fitness function employed by the algorithm prevents irrelevant candidates from taking part in the final solution. Genetic operators are chosen to suit the floating-point representation of the genetic data. As the evolution process improves and the method reaches a near-global solution, a local search is implicitly applied by zooming in on the search interval of each gene by adaptively changing the boundaries of those intervals. The proposed algorithms have produced excellent results in modeling different nonlinear systems with white and colored Gaussian inputs with/without white Gaussian measurement noise

In this paper, fabric faults classification using CNeT (Behnke and Karayiannis, 1998) is studied. The basic objectives are to improve the features selection used in CNeT (Behnke and Karayiannis, 1998) classifier and compare the results with other neural network classifiers. The algorithm adopted here is composed of three stages. The first stage is a preprocessing phase where defects are detected and localized. Since every detected defect has its different shape and size, all defects are normalized to a predetermined size. In the second stage a set of features are calculated for each defect using the Haralick (1973, 1979) spatial features. The improved classification performance is achieved by employing a statistical method to select the most important features that can be used in classification. This is done by calculating a classification factor (Milligan and Cooper, 1985) for each feature vector to determine its effect in the classification process. During the third and last stage, those features are then used to train a competitive neural tree (CNeT) (Behnke and Karayiannis, 1998) designed to learn in a supervised manner the class associated with each set of features. The network can be then used to test and classify new defects. The approach is experimented with a set of images of fault free and faulty textiles and output results are compared with radial basis function classifiers.

This paper presents an FPGA implementation for the Block Truncation Coding (BTC) image compression technique. Images are divided into equal blocks. The BTC calculates the mean of each block for which a two-level quantization is performed so that a zero value is stored for the pixels with values smaller than the mean. The rest of the pixels are represented by the value one. The implementation exploits the inherent parallelism of the algorithm to provide efficient algorithm-to-architecture mapping. FPGA implementation of the BTC is composed of three modules: the input module, to receive input pixels; the quantizer module, to classify pixels to one of the two levels;= and divider circuits to obtain the two quantized values. The implementation is performed for gray and color images. The Xilinx-VirtexE BTC implementation has shown to provide about 23.4x10 6 of pixels/second processing rate which is about 3500 times faster than an Intel Pentium III-550-MHz processor.

The process of training neural networks on parallel architectures has been used to assess the performance of so many parallel machines. In this paper, we are investigating the implementation of backpropagation (BP) on the Alex AVX-2 coarse-grained MIMD machine. A host-worker parallel implementation is carried out in order to train different networks to learn the NetTalk dictionary. First, a computational model is constructed using a single processor to complete the learning process. Also, a communication model for the host-worker topology is developed in order to compute the communication overhead in the broadcasting/gathering process. Both models are then used to predict the machine performance when p processors are used and a comparison with the actual measured performance of the parallel architecture implementation is carried out. Simulation results show that both models can be used effectively to predict the machine performance for the NetTalk problem. Finally, a comparison between the AVX-2 NetTalk implementation and the performance of other parallel platforms is presented.

Given a database of images and their associated data, we want to search the database for a known image so as to retrieve the data associated with that image. The paper proposes a new image-based indexing scheme which enables the retrieval of data associated with an image. The approach proposed is based on using a hierarchy of color moments and offers a compromise between methods based on multiple color moments of segmented regions and those based on global color moments. The method retains positional information better than the above schemes and naturally leads to a multi-level comparison strategy where mismatches are quickly discarded at higher levels. Contrary to other schemes where three moments were used, the proposed methods provide better storage efficiency and less computational efforts as it only employs the first two moments. Simulation results show the efficiency and accuracy of the proposed scheme.

The problem of textile quality control is addressed. The basic objective is to detect faults in woven fabrics. A set of statistical modules is implemented and integrated to detect the faulty textiles. The modules are divided into preprocessing modules (normalization, and median filtering) and processing modules (smoothing, and variance). The algorithms are experimented with a set of images of fault free and faulty textiles and output results are analyzed.

This work presents a new electrocardiogram (ECG) data compression method. By differentiating the signal and using proper thresholding, the ECG is first segmented into a sequence of straight lines. The vertices of these lines are used to encode the signal. The decoding part works by applying the fast orthogonal search (FOS) method (Korenberg and Paarmann 1991) to reconstruct the original signal. Simulation results have demonstrated the efficiency of the algorithm.

Training of feedforward networks on sequential machines is a computationally expensive process. This has motivated the implementation of parallel versions of the backpropagation training algorithm on different parallel platforms in order to decrease the processing time required for training. In this paper, we are investigating the implementation of backpropagation on the Alex AVX-2 coarse-grained MIMD machine. A master–slave parallel implementation is carried out for the encoder–decoder benchmark problem. A communication model for the broadcasting/gathering is used to study the effect of using different topologies. Then the performance of the backpropagation algorithms is analyzed for different network sizes and numbers of processors when the nodes are arranged as a pipeline array and in a mesh topology. © 1998 John Wiley & Sons, Ltd.