Vol. 39 · No. 6

The Optimization Selection Problem is widely known in computer science for its complexity and importance. Several approaches based on machine learning and iterative compilation have been proposed to mitigate this problem. Although these approaches provide several advantages, they have disadvantages that can hinder the performance. This paper proposes a hybrid approach that combines the best of machine learning and iterative compilation. Several experiments were performed using different strategies, metrics and hardware platforms. A thorough analysis of the results reveals that the hybrid approach is a considerable improvement over machine learning and iterative compilation. In addition, the hybrid approach outperforms the best compiler optimization level of LLVM.

Given various features of a software project, it may face different administrative challenges requiring right decisions by software project managers. A major challenge is to estimate software development cost for which different methods have been proposed by many researchers. According to the literature, the capability of a proposed model or method is demonstrated in a specific set of software projects. Hence, the aim of this study is to present a model to take advantage of the capabilities of various software development cost estimation models and methods simultaneously. For this purpose, a new model called ``open hybrid model'' was proposed based on the firefly algorithm. The proposed model includes an extensible bank of estimation methods. The model also includes an extensible bank of rules to describe the relation between existing methods. Considering project conditions, the proposed model tries to find the best rule for combining estimation methods in the methods bank. Three datasets of real projects were used to evaluate the precision of the proposed model, and the results were compared with those of other 11 methods. The results were compared based on performance parmeters widely used to show the accuracy and stability of estimation models. According to the results, the open hybrid model was able to select the most appropriate methods present in the methods bank.

A new method for explaining the Siamese neural network (SNN) as a black-box model for weakly supervised learning is proposed under condition that the output of every subnetwork of the SNN is a vector which is accessible. The main problem of the explanation is that the perturbation technique cannot be used directly for input instances because only their semantic similarity or dissimilarity is known. Moreover, there is no an ``inverse'' map between the SNN output vector and the corresponding input instance. Therefore, a special autoencoder is proposed, which takes into account the proximity of its hidden representation and the SNN outputs. Its pre-trained decoder part as well as the encoder are used to reconstruct original instances from the SNN perturbed output vectors. The important features of the explained instances are determined by averaging the corresponding changes of the reconstructed instances. Numerical experiments with synthetic data and with the well-known dataset MNIST illustrate the proposed method.

The one-sided block Jacobi method (OSBJ) has attracted attention as a fast and accurate algorithm for the singular value decomposition (SVD). The computational kernel of OSBJ is orthogonalization of a column block pair, which amounts to computing the SVD of this block pair. Hari proposes three methods for this partial SVD, and we found through numerical experiments that the variant named ``V2'', which is based on the Cholesky QR method, is the fastest variant and achieves satisfactory accuracy. While it is a good news from a practical viewpoint, it seems strange considering the well-known instability of the Cholesky QR method. In this paper, we perform a detailed error analysis of the V2 variant and explain why and when it can be used to compute the partial SVD accurately. Thus, our results provide a theoretical support for using the V2 variant safely in the OSBJ method.

During the past decades, numerous memory models have been proposed, which focused mainly on how spoken words are studied, whereas models on how visual images are studied are still limited. In this study, we propose a probabilistic memory model (PMM) for visual images categorization which is able to mimic the workings of the human brain during the image storage and retrieval. First, in the learning phase, the visual images are represented by the feature vectors extracted with convolutional neural network (CNN) and each feature component is assumed to conform to a Gaussian distribution and may be incompletely copied with a certain probability or randomly produced in accordance to an exponential distribution. Then, in the test phase, the likelihood ratio between the test image and each studied image is calculated based on the probabilistic inference theory, and an odd value in favor of an old item over a new one is obtained based on all likelihood values. Finally, if the odd value is above a certain threshold, the Bayesian decision rule is applied for image classification. Experimental results on two benchmark image datasets demonstrate that the proposed PMM can perform well on categorization tasks for both studied and non-studied images.

This paper identifies and analyses some drawbacks of the logistic map which is still one of the most used chaotic maps in image encryption algorithms. As some of the disadvantages are caused by inappropriate implementations of the logistic map, this paper proposes a set of rules which should lead to enhancement of the desired chaotic behavior. Probably the most important rule introduces alternating value of parameter utilized by the logistic map. With careful choice of values and an adapted quantization technique, some of the issues should be fixed and theoretically also the values of numerical parameters should be improved. These assumptions are verified by applying the proposed set of rules on an algorithm from our prior work. Effects of the proposed rules on the used algorithm are investigated and all necessary modifications are thoroughly discussed. The paper also compares obtained values of commonly used numerical parameters and computational complexity with some other image encryption algorithms based on more complex chaotic systems.

Images play an important role in modern internet communications, but not all of the images shared by the users are appropriate, and it is necessary to check and reject the inappropriate ones. Deep neural networks do this task perfectly, but it may not be necessary to use maximum power for all images. Many easier-to-identify images may be classified at a lower cost than running the full model. Also, the pressure on the system varies from time to time, so an algorithm that can produce the best possible results for different budgets is very useful. For this purpose, a deep convolutional neural network with the ability to generate several outputs from its various layers has been designed. Each output can be considered as a classifier with its own cost and accuracy. A selector is then used to select and combine the results of these outputs to produce the best possible result in the specified time budget. The selector uses a reinforcement learning model, which, despite the time-consuming learning phase, is fast at execution time. Our experiments on challenging social media images dataset show that the proposed model can reduce the processing time by 32

Recently a new family of five times extended Reed Solomon codes constructed over certain finite fields GF(2 zeta), where zeta >= 3 is an odd integer, was discovered. Until now only an erasure decoding algorithm for these codes was published. In this paper a new decoding algorithm is presented, which allows correcting up to two errors in a codeword from the five times extended Reed Solomon codes. The proposed decoding algorithm is based on syndrome usage.

Twitter is a popular microblogging and social networking service. Twitter posts are continuously generated and well suited for knowledge discovery using different data mining techniques. We present a novel near real-time approach for processing tweets and detecting events. The proposed method, Multi Timing Chained Windows (MTCW), is independent of the language of the tweets. The MTCW defines several Timing Windows and links them to each other like a chain. Indeed, in this chain, the input of the larger window will be the output of the smaller previous one. Using MTCW, the events can be detected over a few minutes. To evaluate this idea, the required dataset has been collected using the Twitter API. The results of evaluations show the accuracy and the effectiveness of our approach compared with other state-of-the-art methods in the event detection in Twitter.