Ongoing PhD Researchers

TOPIC: Attention based Visual Saliency Model for Object Detection

Visual saliency-based algorithms will help to focus on the key parts of a scene. Advances in deep learning and large-scale annotated data resulted in many visual saliency models, but models still fall short in reaching human-level accuracy. A better performance close to human observers in computer vision using visual saliency by incorporating attention mechanisms at behavioral and neural level is proposed. The proposed framework will help the observer to act accordingly to the prediction.

Topic: Design and Analysis of Dynamic Graph Neural Network

Graph Neural Networks(GNNs) have shown their superior ability in learning representations for graph structured data, which leads to performance improvements in many graph related tasks. Most of the existing graph neural network models have been designed for static graphs, while many real-world graphs are inherently dynamic with new nodes and edges constantly emerging. The development of graph neural networks (GNNs) capable of learning representation from graph structured data has led to increased interest in learning the graph representation of the brain connectome. The proposed system aims to develop a Graph Neural Network model for dynamic graph and exploring the applicability of Graph Neural Network in brain connectomics.

  1. Presented a paper titled” Content-based image retrieval approach using convolutional neural network” in the National Conference on Deep Learning and Application(Deep.Learn 2021) in February 2021.
  2. Presented a poster titled” Content-based image retrieval using convolutional neural network” in the third international Conference on Computing and Network communication(CoCoNet’19) in December 2019.
Topic : An Articulation Disorder studies on children with Hearing Disability using Automatic Speech Recognition 

Automatic Speech Recognition (ASR) is defined as the process of converting speech Utterances into corresponding sequences of words by means of computer programs. It aims at understanding and comprehending what was spoken. ASR includes the extraction and determination of the acoustic feature, the acoustic model, and the language model. The extraction and determination of the acoustic feature is a significant part of speech recognition.

Articulation is the process of producing, or making, speech sounds is achieved by controlling the flow of air using articulatory organs. Any sort of obstruction in this articulation process will lead to Speech Articulation Disorder. The proposed work focuses on the identification of speech articulation disorder using Automatic Speech Recognition. This will serve as an aid for diagnosis of Speech Disorder.

 Awarded the National Fellowship for Persons with Disabilities[NFPwD] 2021-22 scholarship.

  1. Presented a paper"titled "Effect of Teaching Digital Electronics to the Deaf Adults through Computer Games" in the International Conference on "Teaching and Rehabilitation Strategies for the Deaf(TRSD)" organized by Anushruti Academy of the Deaf, IIT Roorkee on 24th and 25th September 2022.
  2. Contributed a chapter titled "How Accessibility Changed My Life?" for the book named "Perspectives: Assistive Technology in Limited Resource Settings"[Page No:347-351 published by National Institute of Speech and Hearing [NISH] and Social Justice Department in 2022.
Topic: Speech Recognition using Brain Computer Interface (BCI)

Communication is an act that allows one to share one's emotions with another. Using widely understood words (audio), normal people can communicate easily with them (visual). However, certain cases are preferable for unspoken speech.There are individuals who can't express speech because of physical inabilities which contain traumatic brain injury, brainstem infarcts, stroke as well as ALS (amyotrophic lateral sclerosis) etc. In such situations Brain-Computer Interfaces (BCI) has a growing interest to control external devices or applications using brain waves.This study  focuses on BCI research on the identification of imagined speech.

Topic:Attention and Affect detection in Learning, Multimodal systems

PhD Completed Students

PhD Thesis: Sequential Pattern Matching in DNA Sequence using Logical Match

Abstract: String matching is a method of finding a pattern with some properties within a given sequence of symbols. Much progress has been made over the last few decades in the applications of string matching especially in pattern matching and Computational Biology. Pattern matching algorithms developed for the comparison of molecular sequences are based on the concept of string matching. This thesis presents one such method to locate and compare the sequential pattern using logical match. The method is described in the context of the real world problem, the biological sequence pattern matching. Using the concept of logical match, it is possible, i) to locate repeating DNA sequential pattern ii) to do alignment- free comparison of sequential pattern of finite length.

The thesis demonstrates the application of the method by locating the exact positions of the repeating DNA sequence pattern in the text sequence of the trinucleotide repeat diseases such as Friedreich’s ataxia, Huntington’s disease, Fragile XA syndrome, Dentatorubral- pallidoluysian atrophy (DRPLA), Kennedy’s disease, Myotonic dystrophy- 1, Myotonic dystrophy- 2, Spinocerebellar ataxia type 7 (SCA7), Oculopharyngeal muscular dystrophy and Fragile XE syndrome. This method can be applied to develop a way of research in sequence analysis to locate biologically meaningful segments.

In alignment- free comparison of sequential pattern using logical match, the score calculation is based on the fuzzy membership value which generates from the number of matches and mismatches. The fundamental difference of the alignment- free sequence comparison using logical match with the global alignment using dynamic programming is described here. This thesis puts forward a unique method of search for sequential pattern using correlation matrix memory by logical match. The uniqueness of the method with both artificial data and the real data taken from NCBI (National Centre for Biotechnology Information) databank are demonstrated in this thesis.

Period of Research: 2006- 2011

Research publications arise during PhD program:

Sanil Shanker K P, Elizabeth Sherly and Jim Austin, A note on two applications of Logical Matching Strategy, Applied Artificial Intelligence, 25 (2011) 708–720

Sanil Shanker K P, Aaron Turner, Elizabeth Sherly and Jim Austin, Sequential Data Mining using Correlation Matrix Memory. International Conference on Networking and Information Technology (ICNIT), 2010, Manila, IEEE Xplore, (June 2010) 470- 472.

Sanil Shanker K P, Elizabeth Sherly and Jim Austin, An Algorithm for Alignment- free Sequence Comparison using Logical Match, 2nd IEEE International Conference on Computer and Automation Engineering (ICCAE), 2010, Singapore, IEEE Xplore, 3 (February 2010) 536-538.

Elizabeth Sherly and Sanil Shanker K P, A Fuzzy Inference Scheme for SNP Identification. 2nd IAPR International Workshop on Pattern Recognition in Bioinformatics. Singapore. (October 2007).

Elizabeth Sherly and Sanil Shanker K P, An Exact Pattern Matching Algorithm for Nucleotide Sequences, Proceedings of XIX Kerala Science Congress, Cannanore, (Jan. 2007) 918- 920.

Elizabeth Sherly and Sanil Shanker K P, Storage method for recognizing the pattern of biological sequence, 5th International Conference on Bioinformatics and Biotechnology (InCoB2006), New Delhi, (Dec 2006) 62.

Malayalam Parts Of Speech Tagger for Machine Translation

In Natural Language Processing, morphological analysis plays an important role in processing different forms of human language for formal writing and reading. This thesis depicts the computational aspects of Malayalam Language Processing for machine translation, focusing more on Morphological Analysis and Parts Of Speech (POS) tagging. For morphological analysis, Finite State Models such as Suffix Stripping and Paradigm based methods are used. In POS tagging, machine learning algorithms such as Support Vector Machine (SVM) and Trigram-Ngram (TnT) methods are used. The result shows about 87.4% accuracy by showing better result in a combination of suffix stripping and SVM.

The first chapter provides a general introduction about NLP, Malayalam as an agglutinative language and some approaches to machine translation. The second chapter is on Morph Analysis. Various functions of morphology, Morphological evolution and models have been discussed in this chapter. The third chapter introduces the Parts Of Speech Tagging. Different models on POS tagging and analysis of different tagset and its design are extensively studied and a suitable framework for tagset design for Indian Language especially for Malayalam Language has been made. The fourth chapter dealt with Computational Methods for Morphological Analysis and Parts of Speech Tagging. Hidden Markov Models and Viterbi Algorithm for POS Tagging, Support Vector Algorithm are also discussed in this chapter. Implementation, Results and Discussions of Morphological Anayser and Parts of Speech Tagging have been shown in the fifth chapter. The sixth chapter summarizes the key findings of the research and discusses the scope for future research.

Period of research : 2007-2014

Affiliated University : KERALA UNIVERSITY

Title of Research: On the Specifications of Sequence Constraints in UML

Synopsis:Quality and efficiency of source code is the backbone of software robustness. Reducing manual intervention by automating various phases in software engineering using UML and MDA attempted in this thesis paves the way for paradigm shift in Software Industry. The thesis concentrated on object-oriented constrained modeling combined with a scenario-based behavior specification, that provides an intuitive, yet powerful way to formally specify complex, real-time and safety-critical systems in a reasonably accurate manner. A generalized reusable framework for performing model transformation and thereby code generation of UML constrained behavioral models, facilitating the verification and validation of complex software systems, is developed, which was not addressed effectively till date.

The work presents an automated MT approach; model refactoring, that facilitates efficient code generation from UML behavioral models. The proposed framework helps to improve the model precision thereby establishing consistent code generation by embedding the validation and verification constraints into the model and thus reduces redesign costs through early fault identification and avoidance. A formal verification using a graph matching technique is presented to ensure the behavior consistency of the applied model refactoring process. Agent Based Modeling by applying behavioral design patterns in UML sequence diagrams has been implemented supporting the source code generation. This work is completed as full time research with the support of the Senior Research Fellowship, sponsored by ITRA, Media Lab Asia, Ministry of Electronics and Information Technology (MeiTy), Government of India [August 2014- January 2017]; and Speed IT Fellowship, sponsored by Kerala State ITMission, Government of Kerala [July 2010-July 2013].

Period of Research: March 2011 – October 2016

Affiliated University: University of Kerala

Selected Publications on International Journals and Conferences:

⦁ Chitra M.T., and Elizabeth Sherly. "Trajectory Clustering and Modeling Pattern Identification of Missing Vessels in Deep Sea". IETE-ICRA 2016 – International Conference on Robotics and Automation, February 2016.

⦁ Gayatri Menon B., Chitra M.T., and Elizabeth Sherly. "Optimized Ensemble Based Data Classification for Marine Mobile Communication". 14th Annual International Conference on Mobile Systems, Applications, and Services Companion, ACM MobiSys '16 Companion,ACM Digital Library, Singapore, 2016.

⦁ Chitra, M. T., and Elizabeth Sherly. "Verification of Behavior Preservation in UML Sequence Diagrams using Graph Models." Indian Journal of Computer Science and Engineering, 7(4), 2016.

⦁ Chitra M.T., and Elizabeth Sherly. "UML Behavioral Refactoring for the Specification of Complex Software Systems". Research in Computing Science, Vol.03, National Polytechnic Institute, Mexico, 2015.

⦁ Chitra M.T., and Elizabeth Sherly. “Code Generation from UML Refactored Behavioral Models”, Multi Disciplinary Annual Research Conference (MARC) on Trends and Advancements in ICT, School of Computing, University of Kerala, December 2015.

⦁ Chitra M. T., and Elizabeth Sherly. "Refactoring Sequence Diagrams for Code Generation in UML models." Advances in Computing, Communications and Informatics , ICACCI, 2014, Greater Noida, International Conference on. IEEE,pp. 208-212, 2014. (IEEEXplore) [Scopus, DBLP – indexed]

⦁ Chitra M. T., R. Priya, and Elizabeth Sherly. "Time Based Constrained Object Identification in a Dynamic Social Network." Recent Trends in Computer Networks and Distributed Systems Security(CCIS 335, Springer Berlin Heidelberg, , pp. 314-322, 2012.[ ISBN 978-3-642-34134-2] [Scopus, DBLP – indexed].


The research work was persued with SPEED-IT Fellowship of GoK (2010). The work is focussed to investigate on modeling constrained complex systems for failure detection and prediction to develop robust software systems. The three major contributions are : OMT based model driven engineering through Constraint Satisfaction Problem(CSP), a fault detection model using Neuro-fuzzy for representing imprecise constraints and a fault prediction model using an evolutionary technique, Cellular Automata. The first work is failure detection from the UML class diagram annotated with OCL constraints by mapping to an intermediate form of CSP, with which both verifications and validations could be performed. The power of fuzzy logic, together with the training capability of neural network has been exploited, that helps for fine tuning the fault detection system, in the second work. The third work is a fault prediction technique using evolutionary computing technique of Cellular Automata (CA), where both 1-Dimensional Cellular Automata (1-DCA) and 2-Dimensional Cellular Automata (2-DCA) were employed.

A highly complex constrained system of TTPS (Tamilnadu Tuticorin Power System) which is a coal-fired thermal power plant was used for implementing the three models and could establish that the conventional mathematical model could be replaced by the proposed models. The result obtained from prioritized fuzzy inference system yielded an SSE of 0.0037 and prediction using 2-DCA produced 92% accuracy.Period of Research: March 2011 – October 2016

Affiliated University: University of Kerala


⦁ Priya R and Elizabeth Sherly “Fault Detection and Behavioral Prediction of a Constrained Complex System using Cellular Automata”, International Journal of Computational Science and Engineering (IJCSE), Inderscience Publishers ISSN:1742-7185 [Scopus indexed] [In Press]. DOI: 10.1504/IJCSE.2016.10011939

⦁ Priya R and Elizabeth Sherly “Design of an adaptive constrained based neuro-fuzzy controller for fault detection of a Power Plant System”, Indian Journal of Computer Science and Engineering, Engg. Publishers ISSN:2231-3850, Volume 7, Issue 5, Oct-Nov 2016, pp. 208-218.

⦁ Priya R, Chitra M.T. and Elizabeth Sherly. “Categorization Of Social Networks Based On Multiplicity Constraints .” International Journal of Computer Science Issues, Volume 9, Issue 2, No 2, March 2012 ISSN : 1694-0784 [CiteSeer indexed].


⦁ Priya R, Elizabeth Sherly “Fault Diagnosis of a Constrained Complex system using fuzzy logic controller “, IETE-ICRA 2016 International Conference on Robotics and Automation Conference Proceedings, Feb 2016, ISBN : 978-93-80609-35-5.

⦁ Priya R, Elizabeth Sherly “An Object Constrained UML Modeling for a Complex System”, NCRTIT-2014 National Conference on Recent Trends In Information Technology Conference Proceedings, March 2014 ISBN : 978-81-925229-5-1 pp.23-28.

⦁ Chitra M. T., R. Priya, and Elizabeth Sherly. "Time Based Constrained Object Identification in a Dynamic Social Network." Recent Trends in Computer Networks and Distributed Systems Security(CCIS 335, Springer Berlin Heidelberg, 2012. ISBN 978-3-642-34134-2 pp. 314-322 [Scopus, DBLP – indexed].

Research Title : Texture based Analysis and Classification of Lesions in Medical Images

Synopsis Texture based Analysis and Classification of Lesions in Medical Images

The Breast cancer is one of the most deadly and frequently diagnosed cancers in women. Here an automates system is developed, that uses computer and image processing techniques, combined with machine learning and data mining techniques, for the diagnosis of breast DCE-MRI. The work is focused on two major techniques used for malignancy detection. The first one is the temporal and spatial resolution technique, which is used in Dynamic Contrast Enhanced - Magnetic Resonance Imaging (DCE-MRI) images, that help to study the behaviour of the lesions using lesion enhancement kinetics and the system shows an accuracy of 93.65%. The second one is the structural property, which provides the shape and margin features of lesions in DCE-MRI. A new shape and margin descriptor, circular mesh based shape and margin descriptor (CMSMD), is proposed, feature extraction techniques are derived using the circular mesh based cell labeling techniques and a new algorithmic methodology for the identification of concave and convex regions are developed which shows an accuracy of 94.69% in malignancy detection.

She got Speed IT research fellowship, sponsored by Kerala State IT Mission, Government of Kerala [June 2012 –June 2016] and Senior Research Fellowship, sponsored by ITRA, Media Lab Asia, from Ministry of Electronics and Information Technology (MeiTy), Government of India [May 2017- till date]2. Period of research: December 2012 to November 2016

Affiliated University : University of Kerala

Papers Published:

⦁ Malu, G., Elizabeth Sherly and Sumod Mathew Koshy. "An automated algorithm for lesion identification in dynamic contrast enhanced MRI", International Journal of Computer Applications in Technology , Inderscience 51(1), pp: 23 - 30, 2015.

⦁ Malu, G., Elizabeth Sherly and Sumod Mathew Koshy, “A hybrid approach for Automatic Lesion detection in DCE-MRI of breast using Statistical and Kinetic Features” , Proceedings on the IETE International Conference on Robotics and Automation (ICRA-16), January, 2016. Print ISBN: 978-93-80609-35-5.

⦁ Malu G, Elizabeth Sherly, “A study on different feature extraction techniques for lesion identification in MRI breast images”, Indian Journal of Computer Science and Engineering (IJCSE), ISSN : 0976-5166 Vol. 7 No. 5 Oct-Nov 2016 : 189- 202

⦁ Malu, G., K. Balakrishnan, and N. K. Bodhey. "Area and volume calculation of necrotic tissue regions of heart using interpolation", Proceedings on the 2011 International Conference on Emerging Trends in Electrical and Computer Technology, pp: 728 - 730, IEEE Digital Explorer, 2011. DOI: 10.1109/ICETECT.2011.5760213

⦁ Rugma Mohan, Malu G, Gopakumar, Elizabeth Sherly, “Positioning Lesion from Breast MRI and Mammogram using Registration Method”, International Journal of Engineering Research & Technology. Vol.3(5), May. 2014.

⦁ Lekshmi.B.G, Malu.G, Elizabeth Sherly, “Analysis of Algorithms for the Segmentation of Necrotic Tissues in Cardiac MRI”, Proceedings of the UGC Sponsored National Conference On Recent Trends In Information Technology, 2014. Print ISBN - 978-81-925229-5-1.

google citation :

Research title : Identification of Micro-Calcification In Mammogram Images Using Intelligent Techniques

In this work, we concentrated on the algorithmic development of automated noise removal, contrast enhancement, pectoral muscle removal, feature extraction, segmentation of Region of Interest (ROI), segmentation of micro-calcification clusters from the segmented ROI, feature selection and classification of mammograms. For denoising the mammogram image, modified robust outlyingness ratio-extended non-local means filter (MROR-ENLM) algorithm is employed. A new tracking algorithm integrated with connected component labeling (NTAICCL) is used to remove pectoral muscle (PM) region as well as all other existing artifacts present is proposed. For segmenting ROI that contains micro-calcification clusters in the preprocessed mammograms, a novel feature based spatial fuzzy c-means clustering (FBSFCM) method is implemented. The micro-calcification clusters in the ROI are identified using adaptive h-domes transformation with a threshold based on the intensity classification of the image, which enabled for better classification. The FBSFCM method is selected with mean accuracy of 96.81% and it outperforms other existing methods.

For classification we have developed a new technique using the concept of fuzzy soft set theory to distinguish malignant micro-calcification pattern from a normal or benign one. The advantage of this mathematical procedure is higher power of discrimination and a well determined final solution. Three methods using fuzzy soft set are implemented for the classification of mammogram images as malignant, benign or normal. First method used fuzzy aggregation operation with single feature set and malignant and resulted an accuracy of 94.72%. In the second method, we used fuzzy inter section to combine three separate set of features selected from the images, resulted with an accuracy of 96.58%. Since some valuable information may be lost due to the intersection operation, in the third method we have taken the product operation for finding the final selection and give more accurate result than all other methods. The experiment is performed with 322 images from the MIAS database and resulted in 98.13% accuracy.

Period of Research : March 2011- December 2016

Affiliated University : Kerala University

Papers Published:

⦁ Sreedevi S and Elizabeth Sherly, “Fuzzy Soft Set Approach for Classifying Malignant and Benign Breast Tumors “, Int. J. of Advanced Intelligence Paradigms (IJAIP),2017(accepted Paper).

⦁ S. Sreedevi and Elizabeth Sherly, “A new and efficient approach for the removal of high density impulse noise in mammogram”. International Journal of Computer Aided Engineering and Technology (IJCAET)(Communicated)

⦁ S. Sreedevi and Elizabeth Sherly, "A Novel Approach for Removal of Pectoral muscles in Digital Mammogram," Elsevier Procedia Computer Science, vol. 46, pp. 1726-1731, 2015.

⦁ S. Sreedevi and Elizabeth Sherly, "A New Approach to Mocrocalcification Detection Using Fuzzy Soft Set Approach," Indian Journal of Computer Science and Engineering, vol. 7, no. 2, pp. 46-53, May 2016.

⦁ S. Sreedevi, Terry Jacob Mathew, and Elizabeth Sherly, "Computerized Classification of Malignant and Normal Microcalcifications on Mammograms: using soft set theory", IEEE Xplorer

⦁ S. Sreedevi and Elizabeth Sherly, "A robust approach for denoising of mammographic images contaminated with high density impulse noise," in International Conference on Robotics and Automation, IETE ICRA, 2016, pp. 203-210.ISBN 978-93-80609-35-5

This research work conducted is to develop some optimized methods/ techniques which improve the performance of V Ms (Virtual Machine) in a cloud data center scenario. A lot many factors affect the life cycle of V Ms in a cloud data center. Analyzing all these factors and developing a performance-optimized model for VM management is too complex and difficult to achieve. Hence objective of this research work is to concentrate on dynamic factors which affect the performance of operations on V Ms particularly live migration and load balancing of VMs in cloud data centers. An initial survey conducted to find out the factors affecting the VMs in cloud data centers. Based on the same, the certain algorithms were developed and experimentally analyzed through cloud sim simulator and Xen platform which will improve live migration process, environment of live migration, cost of live migration process, energy efficient load balancing process. Based on these algorithms prototype developed to provide optimized performance in the data center.

List of publication:

  • R. ,Anu, Sherly, Elizabeth, “TAR Based Hotspot Prediction in Cloud Data Centres” International Journal of Grid computing and High Performance Computing, ISSN 19380267 (Under second review process. SCI indexed, Impact factor 0.46, H-index 8)
  • R., Anu, Sherly, Elizabeth, “Performance Optimization Factor Analysis of Virtual Machines Live Migration in cloud Data Centres”, International Journal of Advanced Research in Computer Science, Volume: 8, issue: 5, May 2017.
  • R., Anu, Sherly, Elizabeth, “Live Migration of Delta Compressed Virtual Machines using MPTCP”, International Journal of Engineering research in computer science and engineering, Volume: 4, issue: 9, September 2017.
  • R., Anu, Sherly, Elizabeth, “Simulated annealing Load Balancing for Cloud Data Centres through Ant Colony Optimization”, International Journal of Engineering research in computer science and engineering(Submitted)
  • R., Anu, Sherly, Elizabeth, “Optimized Delta Compression in live Migration of Virtual Machines”, International IEEE Conference on Energy, Communication, Data Analytics and Soft computing (ICECDS), August 2017.
  • R., Anu, Sherly, Elizabeth, “IALM: Interference Aware Live Migration Strategy for Virtual Machines in Cloud Datacentres “, International Springer Conference on Data Management, Analytics and Innovation (ICDMAI), January 2018.
  • R., Anu, Sherly, Elizabeth, “Cost Evaluation of Virtual Machine Live Migration through Bandwidth Analysis”, International Springer Conference on Big data and Cloud Computing (ICBDCC), March 2018.
Title: A Method of Web Document Clustering Using Concept basedMining Model

When the whole universe is shifting to digitization and information becoming unmanageable we resort to assistance of Machines. Information Retrieval using machine learning has been widely accepted technique for effective retrieval of information. The current scenario of WWW widely needs a mechanism that could capture information from diverse distributed environments. The existing methodologies including use of ontologies fails with huge increase of these dynamic distributed patterns exhibited by the information overloads. The changing demands and incremental nature of information requires a methodology that could adapt itself to these dynamic distributed universes of information around us.

Our research interest is focused upon ways to effectively and efficiently cluster documents meeting the dynamic and distributed needs of the search query. For this we have devised a base hybrid algorithm using Frequent Pattern Growth and Fuzzy Particle Swarm Optimization and extended its use to showcase its performance to solve the Short Texts Corpora problem. Also a variant of the proposed algorithm in dynamic environments and in cases to deal with data less than 64MB in Hadoop Architecture has also been modeled.

Finally to prove its efficacies in Recommender System the base algorithm was modified to suit to the needs of the RecSys with more reliability and accuracy.

Book Chapter

  • Pamba R.V., Sherly E., Mohan K .(Accepted).”Self Adaptive Frequent Pattern Growth based Dynamic Fuzzy Particle Swarm Optimization for Web Document Clustering”. Advances in Intelligent Systems and Computing, Springer.

  • Pamba R.V., Sherly E., Mohan K. (2017).” Evaluation of Frequent Pattern Growth Based Fuzzy Particle Swarm Optimization Approach for Web Document Clustering”. In: Gervasi O. et al. (eds) Computational Science and Its Applications – ICCSA 2017. ICCSA 2017. Lecture Notes in Computer Science, vol 10404. Springer, Cham.

    Journal Publications

  • Pamba R.V., Sherly E., Mohan K. (2017). “Automated information retrieval model using FP Growth based fuzzy particle swarm optimization”. International Journal of Computer Science & Information Technology (IJCSIT) Vol 9, No 1, February 2017.

Title: A Dynamic Data Replication Strategy in a Cloud Workflow Environment
  • The emergence of cloud computing technologies offers a new way to develop workflow systems. Cloud computing systems can provide high performance and massive storage required for workflow applications with a lower infrastructure construction cost. In cloud workflows, large amounts of application data need to be stored in distributed data centers. To effectively store these data, a data manager must intelligently select data centers in which these data will reside. When one task needs several data-sets located in different data centers, the movement of large volumes of data becomes a challenge. Data replication is a solution to provide maximum availability of data in which the data is kept closer to the user. This makes the data access efficient and fast with low bandwidth consumption, increased fault tolerance, improved scalability and response time. The data used in workflow applications are usually very large and as such, it is not efficient to replicate all the application data in the system. However, replication of frequently used data reduces the data movement. A good replication strategy can guarantee the security of application data, and reduce the system cost by replicating frequently used data in different locations. The objective is to formulate a distributed data replication strategy using the dynamic data replication which is more efficient, fault-tolerant, secure, consistency preserving and cost-effective than the current strategies.


  • Gopinath, Suji, and Elizabeth Sherly."A Weighted Dynamic Data Replication Management for Cloud Data Storage Systems" International Conference on Computational Intelligence and Data Science (ICCIDS 2018).

  • Gopinath, Suji, and Elizabeth Sherly. "A Weighted Dynamic Data Replication Management for Cloud Data Storage Systems." International Journal of Applied Engineering Research 12.24 (2017): 15517-15524.

  • John, Franklin, Suji Gopinath, and Elizabeth Sherly. "An Efficient Dynamic Data Replication for HDFS using Erasure Coding." International Journal of Computer Science and Information Technologies, 8.2 (2017) , 153-158

  • Franklin John, Suji Gopinath, Elizabeth Sherly. “A decentralised framework for efficient storage and processing of big data using HDFS and IPFS”, (IJHT), International Journal of Humanitarian Technology, Inderscience Publishers. - Accepted

Title: Automatic Speech Recognition for Malayalam

Speech is one of the important mode communication between Humans.Automatic Speech Recognition (ASR) is a process of communication with the help of machines. Machine can understand our speech and convert the speech to human readable text- commonly known as speech Recognition or Speech-To-Text (STT). Preprocessing, Feature extraction, Acoustic modeling, Language Modeling and decoding are the different steps involved in the development of ASR architecture. After the ASR system is generated we can convert commonly used Malayalam words from speech to text in Malayalam Language itself. Futrther we are planning to develop the system to Divyaang People for making their communication easier. In that level of application we are implementing Deep Neural Network (DNN), Convolutional Neural Network (CNN) on Acoustic Modeling section or on Language Modeling section to get more accurate recognition. We are calculating the Word Accuracy, Sentence accuracy and Word Error Rate (WER) for the measurement of ASR system.

Awarded with Kerala State Council for Science, Technology and Environment (KSCSTE) Back-to-Lab fellowship for the project “An assistive Communication Technology for Divyaang” during 2017-2020.


Automatic Speech Recognition using different Neural Network Architectures – A Survey 2016 International Journal of Computer Science and Information Technologies Vol. 7 (6) , 2016, 2422-2427.