THE CSI JOURNAL ON COMPUTER SCIENCE AND ENGINEERING
Year:2005 | Volume:3 | Issue:3 (A)|Papers Count:9

An immune system is an exciting and efficient computational system with many applications in engineering, especially in intrusion detection systems. It is agent-based, distributed and self-adaptive and works based on the hierarchical layered architecture. In this paper, an agent-based artificial immune system based on computational intelligent techniques such as fuzzy logic and genetic algorithms is proposed for computer networks security. The proposed method uses the existing fuzzy relations between antigens and antibodies. Furthermore, it uses the genetic algorithms for optimizing antibodies. To simulate a typical network and to implement network attacks, the network simulator 2 (ns2) is used. To evaluate the performance of the proposed method, the DARPA standard data set is used in training and test phases. The proposed approach is compared with Forrest's immune system as a benchmark. Results show the proposed algorithm detects non-self entities at significantly higher rate, and detectors created by proposed algorithm are more diverse, more robust, and make fewer errors in attack detection.

The high ability of humans in speech perception encourages the use of functional finding of brain in speech recognition. In this research, lexical models are created by using the inversion of neural networks and the NLPCA neural networks. These models can increase the phoneme correction rate up to 81%. We increase the correction rate by combining the lexical model with feature parameters. In this way, by two methods of neural network inversion, feature parameters can be improved, resulting in a correction rate of about 82.4%.

Words are pronounced in various ways in continuous speech. So the lexicon of a continuous speech recognition system is better to contain various pronunciations of each word. The accuracy of word recognition will be improved in this way. An automatic method to generate pronunciation variants of words is introduced in this paper. Pronunciation rules are learned by comparing aligned pairs of reference and recognized phonetic transcriptions of words in this method. In addition, some knowledge-based rewrite rules are added to previous list of learned rules. Consequently after statistical pruning of rules, we have used them to generate pronunciation variants of words by applying them to phonemic transcriptions of words. This method has many advantages in comparison to adding pronunciation variants manually, as it takes into account errors of phone recognizer system and computes application likelihood of each pronunciation variant automatically. This method is implemented by using FARSDAT Persian speech corpus. By usage of generated pronunciation variants in the lexicon of SHENAVA a Persian ACSR, an improvement of as high as 3.47% is achieved in words recognition accuracy.

Speaker identification systems are not only used in ordinary environment but are also used in adverse conditions with obtrusive factors. Voice inconformity can decrease the recognition performance because the training and testing environments may be different. In this paper, our objective is to render speaker identification systems robust against noisy and adverse conditions over telephone or internet. Utterances of 50 speakers from telephony FarsDat Speech database were used to evaluate our speaker identification system in noisy conditions. After removing silence from speech, signal to noise ratio of speech files are changed to 5, 10, 15 and 20dB. LPCC, LFCC, MFCC and MFCC coefficients obtained from Relative Autocorrelation Sequence (RAS) were used as speech features. GMM was used to model speakers. MFCCs were evaluated as the best feature among all cepstral features mentioned above. It was observed that removing the first cepstrum coefficient which represents the frame energy improves identification performance for 10.4%. Linear Weighting of cepstral coefficients, Band Pass Liftering, Cepstral Mean Subtraction, Post Filter, Dynamic Cepstral coefficients were also studied for more robustness. Almost all of these methods improve the identification performance. Linear Weighting was the best method among them. Combinations of the above methods were also evaluated. Most of these combinations led to better performances. The best result was obtained when MFCC coefficients with Linear Weighting and delta MFCC coefficients were used simultaneously in a feature vector.

Model checking is a formal method for verifying finite state systems properties. µ-calculus is very expressive fix point logic capable of specifying a wide range of properties of finite state, reactive and concurrent systems. In this paper, we present a new model checking algorithm for linear and a fragment of indexed modal µ -calculus. This algorithm is based on the method of characterization of fixed point temporal logics formulae using automata. We use first recurrence automata for this purpose. Our algorithm is linear time on the size of the system model. The main contributions of this work are the efficiency of the algorithm and the first use of first recurrence automata for µ- calculus model checking.  

In this paper, a framework for static compilation of applications in reconfigurable computing systems is presented. This framework integrates temporal partitioning and physical design to compile reconfigurable designs. A new temporal partitioning algorithm is proposed, which considers the similarity of node pairs and connections between them. A complementary algorithm attempts to increase node pairs similarity by inserting dummy nodes to partitions. This results in reduction reconfiguration overhead time of subsequent configurations for the devices with partial programming capability. In addition, hard and firm modules are used in placement stage. Incremental placement algorithm has been used to reduce the reconfiguration overhead time and overall run time of the application. Another temporal partitioning algorithm is proposed to address the high memory usage of configurations, which partitions the input data flow graph vertically. A compilation tool has been developed according to the proposed framework.

Exploration-exploitation balance through temperature regulation in multiagent reinforcement learning of different task types is studied. Considered tasks are AND-type, OR-type, and their compositions. The presented study shows that, in contrary to AND-type tasks, the temperature should be set high at the beginning of learning of OR-type tasks and be reduced very gradually during the learning. It is also proposed that, the temperature control policy in learning composite tasks is decided based on the ratio of the number of redundant agents in the learning team to the team population. This ratio shows the similarity of composite task to the two main task types. Learned individual knowledge and the team performance in a simulated benchmarking task are employed for analysis of the presented methods.

In this paper, a new method for color image representation is introduced and image compression-indexing system is described. In the proposed indexing approach, color features are extracted from the compressed image. The performance of the proposed method is evaluated using an image database of 1000 images from 10 semantic groups, as well as a number of standard images based on indexing and compression ability. The experimental results suggest the high performance of the proposed method.

The combination of multiple classifiers is shown to be suitable for improving the performance of pattern recognition systems. The theoretical and experimental results in the literature clearly show that combining multiple classifiers is only effective if the individual classifiers are accurate and diverse. Selecting a suitable combination method and the diversity of an ensemble of classifiers are known to be important key issues in constructing a good ensemble system. The combination method should be selected so that the classifiers complement each other. In this paper we review diversity creation methods including implicit and explicit methods. Also a review on combination methods is presented that covers the following rules: maximum, minimum, mean, product, voting, Bayesian, fuzzy integral, decision template and Dempster-Shafer.