Showing papers by "Anupam Shukla published in 2010"

Fusion of probabilistic A* algorithm and fuzzy inference system for robotic path planning

Abstract: Robotic Path planning is one of the most studied problems in the field of robotics. The problem has been solved using numerous statistical, soft computing and other approaches. In this paper we solve the problem of robotic path planning using a combination of A* algorithm and Fuzzy Inference. The A* algorithm does the higher level planning by working on a lower detail map. The algorithm finds the shortest path at the same time generating the result in a finite time. The A* algorithm is used on a probability based map. The lower level planning is done by the Fuzzy Inference System (FIS). The FIS works on the detailed graph where the occurrence of obstacles is precisely known. The FIS generates smoother paths catering to the non-holonomic constraints. The results of A* algorithm serve as a guide for FIS planner. The FIS system was initially generated using heuristic rules. Once this model was ready, the fuzzy parameters were optimized using a Genetic Algorithm. Three sample problems were created and the quality of solutions generated by FIS was used as the fitness function of the GA. The GA tried to optimize the distance from the closest obstacle, total path length and the sharpest turn at any time in the journey of the robot. The resulting FIS was easily able to plan the path of the robot. We tested the algorithm on various complex and simple paths. All paths generated were optimal in terms of path length and smoothness. The robot was easily able to escape a variety of obstacles and reach the goal in an optimal manner.

Breast cancer diagnosis using Artificial Neural Network models

Abstract: Breast cancer is the second leading cause of cancer deaths worldwide and occurrs in one out of eight women. In this paper we develop a system for diagnosis, prognosis and prediction of breast cancer using Artificial Neural Network (ANN) models. This will assist the doctors in diagnosis of the disease. We implement four models of neural networks namely Back Propagation Algorithm, Radial Basis Function Networks, Learning vector Quantization and Competitive Learning Network Experimental results show that Learning Vector Quantization shows the best performance in the testing data set This is followed in order by CL, MLP and RBFN The high accuracy of the LVQ against the other models indicates its better ability for solving the classificatory problem of Breast Cancer diagnosis.

Dynamic environment robot path planning using hierarchical evolutionary algorithms

Abstract: The problem of path planning deals with the computation of an optimal path of the robot, from source to destination, such that it does not collide with any obstacle on its path. In this article we solve the problem of path planning separately in two hierarchies. The coarser hierarchy finds the path in a static environment consisting of the entire robotic map. The resolution of the map is reduced for computational speedup. The finer hierarchy takes a section of the map and computes the path for both static and dynamic environments. Both the hierarchies make use of an evolutionary algorithm for planning. Both these hierarchies optimize as the robot travels in the map. The static environment path is increasingly optimized along with generations. Hence, an extra setup cost is not required like other evolutionary approaches. The finer hierarchy makes the robot easily escape from the moving obstacle, almost following the path shown by the coarser hierarchy. This hierarchy extrapolates the movements of the various objects by assuming them to be moving with same speed and direction. Experimentation was done in a variety of scenarios with static and mobile obstacles. In all cases the robot could optimally reach the goal. Further, the robot was able to escape from the sudden occurrence of obstacles.

Offline Handwriting Recognition using Genetic Algorithm

Abstract: Handwriting Recognition enables a person to scribble something on a piece of paper and then convert it into text. If we look into the practical reality there are enumerable styles in which a character may be written. These styles can be self combined to generate more styles. Even if a small child knows the basic styles a character can be written, he would be able to recognize characters written in styles intermediate between them or formed by their mixture. This motivates the use of Genetic Algorithms for the problem. In order to prove this, we made a pool of images of characters. We converted them to graphs. The graph of every character was intermixed to generate styles intermediate between the styles of parent character. Character recognition involved the matching of the graph generated from the unknown character image with the graphs generated by mixing. Using this method we received an accuracy of 98.44%.

Intelligent decision support system for breast cancer

Abstract: Breast cancer is the second leading cause of cancer deaths in women worldwide and occurs in nearly one out of eight women Currently there are three techniques to diagnose breast cancer: mammography, FNA (Fine Needle Aspirate) and surgical biopsy In this paper we develop an integrated expert system for diagnosis, prognosis and prediction for breast cancer using soft computing techniques The basic aim is to compare the various neural network models from the recent literature Breast cancer database used for this purpose is from the University of Wisconsin (UCI) Machine Learning Repository Three different data sets have been used, each employing different diagnostic technique It can use diagnosis, prognosis and survivability prediction of breast cancer patient in one intelligent system We implement six models of neural networks namely Back Propagation Algorithm, Radial Basis Function Networks, Learning vector Quantization, Probabilistic Neural Networks, Recurrent Neural Network, and Competitive Neural network Experimental Results show that different models give optimal performance for different types of data sets However, all the models are able to solve the problem to a reasonable extent

Analysis of Artificial Neural Network for Financial Time Series Forecasting

Abstract: Financial forecasting has been challenging problem due to its high non-linearity and high volatility. An Artificial Neural Network (ANN) can model flexible linear or non-linear relationship among variables. ANN can be configured to produce desired set of output based on set of given input. In this paper we attempt at analyzing the usefulness of artificial neural network for forecasting financial data series with use of different algorithms such as backpropagation, radial basis function etc. With their ability of adapting non-linear and chaotic patterns, ANN is the current technique being used which offers the ability of predicting financial data more accurately. "A x-y1 network topology is adopted because of x input variables in which variable y was determined by the number of hidden neurons during network selection with single output." Both x and y were changed.

Fusion of Speech and Face by Enhanced Modular Neural Network

Abstract: Biometric Identification is a very old field where we try to identify people by their biometric identities. The field shifted to bi-modal systems where more than one modality was used for the identification purposes. The bi-modal systems face problem related to high dimensionality that may many times result in problems. The individual modules already have large dimensionality. Their fusion adds up the dimensionality resulting in still larger dimensionality. In this paper we solve these problems by the introduction of modularity at these attributes. Here we divide various attributes among various modules of the modular neural network. This limits their dimensionality without much loss in information. The integrator collects the probabilities of the occurrences of the various classes as outputs from these neural networks. The integrator averages these probabilities from the various modules to get the final probability of the occurrence of each class. This averaging is performed on the basis of the efficiencies of the modules at the time of training. A module that is well trained is hence expected to give a better performance than the one which is not well trained. In this manner the final probability vector may be calculated. Then the integrator selects the class that has the highest probability of occurrence. This class is returned as the output class. We tested this algorithm over the fusion of face and speech. The algorithm gave good recognition of 97.5%. This shows the efficiency of the algorithm.

Sequential combination of statistics, econometrics and Adaptive Neural-Fuzzy Interface for stock market prediction

Abstract: Since last decade advanced data simulations help to identify hidden trends in a time series Our purpose is to identify uncertainties during recession period using statistical analysis, econometrical analysis and Adaptive Neural-Fuzzy networks In this paper, initially through computational analysis we are testing financial data using correlation tests, likelihood tests, heteroscedastic characteristics analysis and hypothesis tests These statistical and econometrical tests give us exact nature of data set and relation between data points All tests and analysis are studied on NASDAQ Stock Market over last 2-years Then after, optimized subtractive data clustering method is used to cluster the data and create fuzzy membership functions by using Sugeno-type Fuzzy Interface System (FIS) Finally, we are using optimized hybrid learning algorithm in customized Adaptive Neural Fuzzy Interface System (ANFIS) to train the network Hence, we got an efficient Adaptive Neural-Fuzzy network to check and test the data sets and use it for forecasting the stock market index During this, the hybrid learning algorithm combines Least-Square method and the Back-propagation gradient descent methods for training the Fuzzy Interface System (FIS) with the help of optimized membership functions and parameters This paper presents a state-of-art for Adaptive Neural-Fuzzy Network (ANFN) application to forecast stock market index and involved market uncertainties by combining the econometrical test to optimize the ANFIS and FIS function

Text-Dependent Multilingual Speaker Identification for Indian Languages Using Artificial Neural Network

Abstract: In this paper an attempt is made to develop speaker identification system which is used to determine the identity of an unknown speaker among several speakers of known speech characteristics, from a sample of his or her voice. Every speaker has different individual characteristics embedded in his /her speech utterances. These characteristics can be extracted from utterances and different neural network models are used to get the desired results. To evaluate speech characteristics from utterances they are stored in digitized form. Speech features namely LPC, RC, APSD, Number of zero crossing and Formant frequencies are extracted from speech signal and formed speech feature vectors. These data features are fed into Artificial Neural Network using back propagation learning algorithm and clustering algorithm for training and identification processes of different speakers. The database used for this system consists of 20 speakers including both male and female from different parts of India and languages are Hindi, Sanskrit, Punjabi and Telugu. The average identification rate 83.29% is achieved when the network is trained using back propagation algorithm and it is improved by about 9% and reached up to 92.78% when using clustering algorithm.

Use of Modular Neural Network for Heart Disease

Abstract: The medical field is very versatile field and one of the interested research areas for the scientist. It deals with many medical disease problems starting with the diagnosis of the disease, preventing from the disease and treatment for the disease. There are various types of medical disease and accordingly various types of treatment methods. In this paper we mostly concern about the diagnosis of the heart disease. Mainly two types of the diagnosis method are used one is manual and other is automatic diagnosis which consists of diagnosis of disease with the help of intelligent expert system. In this paper the modular neural network is used to diagnosis the heart disease. The attributes are divided and given to the two neural network models Backpropagation Neural Network (BPNN) and Radial Basis Function Neural Network (RBFNN) for training and testing. The two integration techniques are used two integrate the results and provide the final training accuracy and testing accuracy. The modular neural network with probabilistic product method gave an accuracy of 87.02% over training data and 85.88% over testing accuracy and with probabilistic product method gave an accuracy of 89.72% over training data and 84.70% over testing accuracy, which was experimentally determined to be better than monolithic neural networks.

Multilingual speaker recognition using ANFIS

Abstract: Feature based Recognition Systems has been an area of intense research for long. The creation of a reliable, robust and sufficiently efficient recognition system has been tried using features from several sources including textual and image sources. Speech based sources have also been used for the creation of such a recognition system. However, variations caused due to differences in individual speaker characteristics, mood variations and inter-mingled noise disturbances make the realization of such a system very difficult. This paper proposes a recognition system for identification of the speaker, language and the words spoken. The system makes use of Adaptive Neuro-Fuzzy Inference paradigm for the same. First, the sampling frequency and the speech features are extracted from the speech database to form speech feature vectors. The features used are LPC, LPCC, RC, LAR, LSF and ARSCIN. The speech database is prepared using 25 speakers including male and female speakers. Five different speaking texts of different languages having same meaning are used to get the best speaker identification accuracy. The languages spoken by the speakers include English, Hindi, Punjabi, Sanskrit and Telugu. The Feature vectors, thus prepared, are fed to an Adaptive Neuro-Fuzzy Inference System for speaker, language and word recognition. The experimental results show the system to be amply efficient and successful in the recognition tasks involved.

Diagnosis of breast cancer by modular neural network

Abstract: Diagnosis of diseases is well known problem in the medical field. Past research shows that medical database of disease can be train by using various neural network models. Many medical problems face the problem of curse of dimensionality due to the excessively large number of input attributes. Breast cancer is one such problem. We propose the use of modular neural network for effective diagnosis. In the proposed methodology four modules are made; each module gets half the problem attributes which are trained and tested by two neural network models, Back Propagation Neural Network (BPNN) and Radial Basis Function (RBFN). Integration is done using a probabilistic sum rule. The modular neural network gave an accuracy of 95.75% over training data and 98.22% over testing accuracy, which was experimentally determined to be better than monolithic neural networks.

A Novel Approach to Classificatory Problem using Grammatical Evolution based Hybrid Algorithm

Abstract: Numerous problems that where intelligent systems find application are classificatory in nature. These include Face Recognition, Speaker Recognition, Word Recognition, etc. In this paper we propose a new model for these classificatory problems. This model is based on the Grammatical Evolution which is an Evolutionary Algorithm that uses chromosomes as a set of instructions over a predefined grammar. The model that we propose here is a type of fuzzy inference system. Rules are in form of a collection of points representing every class. The separation between the unknown input and these representative points determines the degree of belongingness of the unknown input to the specific class being considered. Multiple contributions from same classes are simply added together. The training data set is used for the purpose of generating the initial set of configurations of this fuzzy model. The fuzzy functions are parameterized by adding fuzzy parameters, like any neuro-fuzzy model. These parameters are trained by a validation data set using a training algorithm. The performance of the system over training and validation data set serve as the fitness function. Variable mutation rate is applied. We tested the effectiveness of the algorithm over the picture learning problem and received better results than numerous commonly used algorithms. General Terms Algorithms

Hybrid Intelligent Systems for Medical Diagnosis

Abstract: Bio-Medical Engineering is a rapidly growing field as a result of the need and rise of automation. This field calls for the collaboration between the people from the medical background and the engineers to develop intelligent systems for the various tasks in bio-medicals. These systems are used for the detection of the various diseases. These act as Clinical Decision Support Systems (CDSS) in order to assist the doctors in their task of identification of the presence of the diseases. They hence act as valuable tools for the doctors in the analysis of the diseases. This is especially important considering the work load over the doctors and the vast presence of the diseases. The increasing health consciousAbstrAct

Breast cancer diagnostic system using Symbiotic Adaptive Neuro-evolution (SANE)

Abstract: Breast cancer is the second leading cause of cancer deaths in women worldwide and occurs in nearly one out of eight women. In this paper we develop a hybrid intelligent system for diagnosis, prognosis and prediction for breast cancer using SANE (Symbiotic, Adaptive Neuro-evolution) and compare with ensemble ANN, modular neural network, fixed architecture evolutionary neural network (F-ENN) and Variable Architecture evolutionary neural network (V-ENN). While the monolithic neural and fuzzy systems have been extensively used for diagnosis, the individual limitations of the various models put a great threshold on prediction accuracies, which may be overcome with the use of SANE. The SANE system coevolves a population of neurons that cooperate to form a functioning neural network. Breast cancer database from the University of Wisconsin available at UCI Machine Learning Repository is used for conducting experimental work.

Expert System for Speaker Identification Using Lip Features with PCA

Abstract: Biometric authentication techniques such as lips, face, and eyes are more reliable and efficient than conventional authentication techniques such as password authentication, token, cards, personal identification number, etc. In this research paper, the emphasis has been laid on the speaker identification based on lip features. In this study, we have presented a detailed comparative analysis for speaker identification by using lip features, Principal Component Analysis (PCA), and neural network classifiers. PCA has been used for feature extraction from the six geometric lip features which are height of the outer corners of the mouth, width of the outer corners of the mouth, height of the inner corners of the mouth, width of the inner corners of the mouth, height of the upper lip, and height of the lower lip. These features are then used for training of the network by using different neural network classifiers such as Back Propagation (BP), Radial Basis Function (RBF) and Learning Vector Quantization (LVQ). These approaches are incorporated on "TULIPS1 database, (Movellan, 1995)" which is a small audiovisual database of 12 subjects saying the first 4 digits in English. After the detailed analysis and evaluation a maximum of 91.07% accuracy in speaker recognition is obtained using PCA and RBF. Speaker identification has a wide range of applications such as Audio Processing, Medical data, Finance, Array processing, etc.

Sentence Recognition Using Hopfield Neural Network

Abstract: Communication in natural languages between computational systems and humans is an area that has attracted researchers for long. This type of communication can have wide ramification as such a system could find wide usage in several areas. WebBrowsing via input given as textual commands/sentences in natural languages is one such area. However, the enormous amount of input that could be given in natural languages present a huge challenge for machine learning of systems which are required to recognize sentences having similar meaning but different lexico-grammatical structures. In this paper, we describe how a binary recurring neural network can be used to sufficiently solve this problem for English. The system uses the Hopfield Neural Network to recognize the meaning of text using training files with limited dictionary. Detailed analysis and evaluation show that the system correctly recognizes/classifies approximately 92.2% of the input sentences according to their meaning.

Generalized and modified ant algorithm for solving robot path planning problem

Abstract: The task of planning trajectories for a mobile robot has received considerable attention in the research literature. The problem involves computing a collision-free path between a start point and a target point in environment of known obstacles. In this paper, we introduced the generalized and modified ant algorithm for solving robot path planning, by the term generalized we mean ant can select either one of the 16, 24 or 32 neighbor points for its next movement as contrast to simple one in which only one of the eight neighborhoods can be selected by the ant. As per the general theory of the graph search algorithms, the increase in the number of neighborhood points make the solutions more optimal in terms of path length, but put a limitation on the execution time which increases drastically with an increase in the number of neighborhood points. Our simulation results show that there is a considerable decrease in the path length with the increase in the level of generalization, the time of execution however increases but the algorithm performance can be improved by modified ant algorithm in terms of execution time.

Clustering-based hierarchical genetic algorithm for complex fitness landscapes

Abstract: We propose the use of a hierarchical genetic algorithm (GA) for optimisation in complex landscapes. While the slave GA tries to find the local optima in the restricted fitness landscape of low complexity, the master GA tries to identify interesting regions in the entire landscape. The slave GA is a conventional GA with high convergence. The master GA is more exploratory in nature. This GA clusters the fitness landscape with each cluster in control of a slave GA. The number of clusters decreases with time to get global characteristics. The novelty of the suggested approach lies in the trade-off between the search for global optima and convergence to local optima that can be controlled between the two GAs. We tested the algorithm and observed that the approach exceeds conventional GA as well as particle swarm optimisation in complex landscapes.

Multimodal Biometric Systems

Abstract: The uni-modal biometric systems making use of a single biometric modality have a limited performance that restricts their applicability in real life scenarios. The multimodal biometric systems make use of two or more modalities that together achieve much higher performances. In this chapter we discuss the means to fuse three modalities to make a more robust system. We first discuss the fusion of speech, lip, and face. This system uses Hidden Markov Models for the classification and an integration technique called as late integration for decision making from the three modalities. We then discuss the fusion of face, speech and fingerprint. Here each of the individual biometric modalities would make use of modular neural network which would then be combined using a fuzzy integration technique. The last model we discuss would carry the fusion of fingerprint, face and hand geometry. This system uses a variety of fusion techniques including a sum rule, linear discriminant function and decision trees.

Automatic Summary Generation from Single Document Using Information Gain

Abstract: With tons of information pouring in every day over Internet, it is not easy to read each and every document. The information retrieval from search engine is still far greater than that a user can handle and manage. So there is need of presenting the information in a summarized way. In this paper, an automatic abstractive summarization technique from single document is proposed. The sentences in the text are identified first. Then from those sentences segments, unique terms are identified. A Term-Sentence matrix is generated, where the column represents the sentences and the row represents the terms. The entries in the matrix are weight from information gain. Column with a maximum cosine similarity is selected as first sentence of the summary sentence and likewise. Results over documents indicate that the performance of the proposed approach compares very favorably with other approaches.

Modular Neural Networks

Abstract: Modular Neural Networks are use of a number of Neural Networks for problem solving. Here the various neural networks behave as modules to solve a part of the problem. The entire task of division of problem into the various modules as well as the integration of the responses of the modules to generate the final output of the system is done by an integrator. In this chapter we first look at the various Modular Neural Network models. Here we would mainly study two major models. The first model would cluster the entire input space with each module responsible for some part of it. The other model would make different neural networks work over the same problem. Here we would be using a response integration technique for figuring out the final output of the system. The other part of the chapter would present Evolutionary Modular Neural Networks. We would first present a simple genetic approach and then a co-evolutionary approach for this evolution of the entire Modular Neural Network.

Speech Signal Analysis

Abstract: Intelligent systems possess the capability to model and solve many problems of practical importance. The best way to understand these systems is do design and develop such systems which exposes their various advantages and disadvantages. This chapter presents the basic analysis technique of speech signals that would further help us in using speech as a medium of developing intelligent systems. In this chapter we study the manner in which we may highlight and extract useful features out of a given speech signals. We discuss the analysis techniques in two heads. The first head consist of the bank of filters approach. Here we present the Fourier, Short Time Fourier and Wavelet Analysis which extract interesting features. Here we would stress the importance of time and frequency domain. In the other head we would discuss the Linear Predictive Coding (LPC) methods. We discuss the manner in which the linear coding helps in analysis. We even discuss the general speech parameters that facilitate good recognition in these intelligent systems.

Evolving robotic path with genetically optimised fuzzy planner

Abstract: Path planning is one of the highly studied problems in the field of robotics. The problem has been solved using numerous statistical, soft computing and other approaches. In this paper, we evolve the robotic path using genetic algorithms (GA). The GA generates solutions for the static map which disobeys the non-holonomic constraints. Fuzzy inference system (FIS) works on the generated path and extends the problem for dynamic environment. The results of GA serve as a guide for FIS planner. The FIS system was initially generated using rules from common sense. Once this model was ready, the fuzzy parameters were optimised by another GA. The GA tried to optimise the distance from the closest obstacle, total path length and the sharpest turn at any time in the journey of the robot. The resulting FIS was easily able to execute the plan of the robot in a dynamic environment. We tested the algorithm on various complex and simple paths. All paths generated were optimal in terms of path length and smoothness. Hence, using a combination of GA along with FIS, we were able to solve the problem of robotic path planning.

Distributed Growing Radial Basis Function Neural Networks for Location Estimation in Indoor Wireless Networks

Abstract: Location aware computing is popularized and use of location information has become important due to huge application of mobile computing devices and local area wireless networks. To serve us well, mobile computing applications need to know the physical location of things so that they can record them and report them to us. Therefore in the future ubiquitous services, location estimation will be a key technology. This paper present distributed growing radial basis function neural networks (DGRBFNN) for location estimation of mobile device in wireless networks. DGRBFNN partitions the location space into clusters on the basis of availability of signals from various access points, and employ separate neural network architecture for each cluster to estimate the location of mobile device in indoor wireless networks. It provides better location estimation results than other approaches and systematically caters for unavailable signals at estimation time.

A Fast Mode Selection Algorithm Using Texture Analysis for H.264/AVC

Abstract: H.264/AVC, the newest international video coding standard achieves higher compression efficiency as compared to all other existing coding standard such as MPEG-4 and H.263. However this efficiency comes with a dramatic increase in computational complexity due to several advanced techniques, such as inter mode and intra mode prediction with variable size motion compensation. It adopts rate distortion optimization (RDO), while maximizing visual quality and minimizing the required bit rate. In this paper, we propose a fast inter mode selection algorithm. The aim is to reduce the number of modes in intra and inter mode prediction. Experimental results show that this algorithm reduces the total encoding time with little loss of bit rate and visual quality.

Artificial Neural Networks

Abstract: Artificial Neural Networks (ANN) are an inspiration from the human brain. These systems contain a large number of neurons that work in a parallel architecture. Each neuron takes its input directly from system or from other neurons. The information is processed and given to the other neurons. This is the basic phenomenon that makes possible all simple and complex problem solving ability of these networks. The chapter discusses the various models of neural networks that include multi-layer perceptron with back propagation algorithm, radial basis function networks, learning vector quantization, self organizing maps and recurrent neural networks. We discuss the basic philosophies and problem solving approach of these networks. A lot of emphasis is given on the various system parameters and their role and importance in the overall system design. We further illustrate the various limitations of the different models. This forms the motivation behind the use of hybrid systems that we present in the subsequent chapters.