Showing papers by "Anupam Shukla published in 2016"

Multi robot area exploration using nature inspired algorithm

Abstract: Exploration is a process where the objective is to cover an area that is used for subsequent navigation. It is an important criteria for problem-solving in many unknown search space and is an important aspect of automation and information gathering. Here we have proposed multi-robot area exploration method for unknown search areas. In the proposed work, exploration is mainly guided by combined effect of probabilistic and deterministic movement. It uses Clustering Based Distribution Factor (CBDF) for deterministic movement and nature inspired algorithms (NIA) like Particle swarm optimization (PSO), Bacteria foraging optimization (BFO), and Bat algorithm (BA) for random guidance for exploration. The environment partitioning avoids repeated area coverage, and robots may be allocated to any partition to explore the map in a random manner. Robots move in the direction provided by CBDF and explore the area using nature inspired algorithm. The proposed approaches have been implemented and evaluated in several simulated environments and with varying team sizes and detection ranges. Simulation results show that, on increasing the number of robots and detection range, performance also increases and that best results are achieved for PSO.

Soft computing based expert system for Hepatitis and liver disorders

Abstract: The main objective of this research work is to develop an expert system for the diagnosis and detection of Hepatitis and liver disorders based on various Artificial Neural Networks models. In this research work Artificial Neural Networks models like Back Propagation Algorithm, Probabilistic Neural Networks, Competitive learning Networks, Learning vector quantization and Elman Networks have been used for detection and diagnosis of Hepatitis and liver disorders. The various networks developed with the help of MATLAB. Required data has been chosen from trusty machine learning data base (UCI). This system in comparison with other traditional diagnostic systems is faster, more reliable and more accurate. One can use this system as a specialist assistant or for training medicine students.

Improving microgrid voltage stability through cyber-physical control

Abstract: Voltage stability in islanded microgrids can be improved through coordinated cyber-physical control among the distributed generators(DGs). This paper proposes two techniques for developing controllers for handling situations of varying loads and communication link failure-one for a full communication topology type microgrid and the other for a sparse communication type. In the first technique, the working of the microgrid is represented as set of switching systems and controllers are developed using the Common Lyapunov Function theory. In the second technique a specially designed constraint based sensor-controller connection design (CBSCD) algorithm has been adopted to manipulate the communication topology and improve stability in response to varying parametric conditions. The algorithm finds the least set of communication resources to be operational while enhancing the stability of the microgrid voltage to a near maximal level. Both the methodologies have been tested for an islanded 4 bus power system for varying loads and communication link failure.

Parkinson's Disease Detection with Gait Recognition using Soft Computing Techniques

Abstract: Parkinson’s disease is a degenerative disorder of the central nervous system which occurs as a result of dopamine loss, a chemical mediator that is responsible for body’s ability to control the movements. It’s a very common disease among elder population effecting approx 6.3 million people worldwide across all genders, races and cultures. In this chapter, authors have proposed an automated classification system based on Artificial Neural Network using Feed Forward Back-propagation Algorithm for Parkinson’s disease diagnosis by analyzing gait of a person. The system is trained, tested and validated by a gait dataset consisting data of Parkinson’s disease patients and healthy persons. The system is evaluated based on several measuring parameters like sensitivity, specificity, and classification accuracy. For the proposed system observed classification accuracy is 97.11% using 19 features of gait, and 95.55% using 10 prominent features of gait selected by Genetic Algorithm.

Wave Front Method Based Path Planning Algorithm for Mobile Robots

Abstract: Path planning problem revolves around finding a path from start node to goal node without any collisions. This paper presents an improved version of Focused Wave Front Algorithm for mobile robot path planning in static 2D environment. Existing wave expansion algorithms either provide speed or optimality. We try to counter this problem by preventing the full expansion of the wave and expanding specific nodes such that optimality is retained. Our proposed algorithm ‘Optimally Focused Wave Front algorithm’ provides a very attractive package of speed and optimality. It allocates weight and cost to each node but it defines cost in a different fashion and employs diagonal distance instead of Euclidean distance. Finally, we compared our proposed algorithm with existing Wave Front Algorithms. We found that our proposed approach gave optimal results when compared with Focused Wave Front Algorithm and faster results when compared with Modified Wave Front Algorithm.

Accoustic Modeling for Development of Accented Indian English ASR

Abstract: This paper investigates Indian English from the point of view of a speech recognition problem. A novel approach towards building an Automated Speech Recognition System (ASR) for Indian English using PocketSphinx has been proposed. The system was trained with a database of English words spoken by Indians in three different accents using continuous as well as semi-continuous models. We have compared the performances in each case and the optimum case performance comes close to 98 % accurate. Based on this study, we tweaked the original PocketSphinx Android application in order to incorporate our results and present it as an Indian English-based SMS sending application. We are working further on this approach to identify ways of successfully training a speech recognition system to recognize a much wider variety of Indian accents with much more significant accuracy.

Expert system for breast cancer diagnosis using ensemble approach

Abstract: Development of efficient, prompt and robust systems that are intelligent enough to replace/reduce human supervision in medical diagnosis is one of the primary objectives that have driven advancements in research for long. One area where this need for intelligent automation has been most acutely felt is related to the diagnosis of breast cancer in women. Mammography is the most widely used test for screening and early diagnosis of breast cancer. However, it is error-prone and hence cannot be used reliably for effective diagnosis of the said disease. In this paper, we describe how the above said problem could be efficiently solved through Ensemble Approach. The use of the approach has bestowed the expert system with significantly simple and swift learning, smaller requirement for storage space during classification, faster classification with added possibility of incremental learning. The system was comparatively evaluated using different ensemble integration methods for breast cancer diagnosis namely weighted averaging, product, minimum, maximum, poling and fuzzy integration and different neural network approaches including MLP neural network, Radial Basis Function Network, Learning Vector Quantization and Recurrent Neural Network. Detailed experimental analysis with the system shows that the best performance in terms of accuracy and specificity measures is achieved while used maximum integration technique with Radial Bass Function Network while finest performance in terms of sensitivity is achieved when MLP neural network with Minimum integration is used.

Acoustic modeling using state projection vectors of subspace Gaussian mixture model to train deep neural network on entropy maximized Hindi dataset

Abstract: Recent advancements and efficient training procedures in deep neural networks (DNNs) have significantly outperformed the hidden Markov model-Gaussian mixture model (HMM-GMM). The performance of DNNs can further be improved should it be given better phonetic context information. This is manifested by state specific vectors (SSV) of subspace Gaussian mixture model (SGMM). In this paper, we use the state specific vectors of SGMM as features to provide additional phonetic context information to the DNN framework. The state specific vectors are aligned with each observation vector of the training data to form the state specific vector (SSV) feature set. The combination of linear discriminant analysis (LDA) feature sets and state specific feature sets are then used as input features to train the DNN framework. Relative improvement of up to 4.13% is obtained on Hindi database using DNN trained with a combination of state specific feature sets and LDA feature sets compared to the DNN trained only with LDA feature sets. Since state specific vectors provide extra information about the phonetic context, they show improved results when combined with DNN framework. In this paper, we also investigate the performance of speech recognition on different training data selection strategies. The idea is to implement an approach that maximizes the information content in the training corpus. The experiments in this paper are carried on the training data set having maximum information content.

Improved Identity Anonymization Using Hashed-TOR Network

Abstract: At present, the modern world has been computerized to a huge extent because of the increasing penetration of the Internet There is a plethora of services available on the web platform these days Services available may be either legitimate or illegitimate International borders have been effectively eliminated with the help of the internet and so more and more companies are offering their services across the globe The physical location of the service provider does not matter anymore Since every request that is sent to the WWW, every message that is relayed and, in fact, every click is logged somewhere on the internet, it gives rise to a huge knowledge base of user data This knowledge base can be manipulated and exploited in a number of ways This poses a great threat to any individual user who wants to utilize the services available on the internet anonymously A mechanism is required using which individuals can maintain their own privacy from global organizations such as the NSA, Google, etc, which are believed to collect huge amounts of personal user data The TOR network goes a long way in doing that and abstracts the user from the conventional internet But TOR itself is not free from vulnerabilities In this paper, a vulnerability with the ONION routing protocol, which is the spine of the TOR network, has been presented It has been analyzed and a technique to overcome it has also been discussed to make the TOR network safer for anonymous use on the World Wide Web