Computer Science Engineering Research Projects

The School is actively involved in cutting edge research encompassing the areas of Cloud Computing, Data Mining, Image Processing, Network Security and Wireless Sensor Networks. The School has also made significant contributions towards knowledge generation by presenting/publishing papers in National and International journals/Conferences through work of faculty members and UG/PG students.

On Going Computer Science Research Activities

The School of Computer Science & Engineering is actively involved in cutting edge research encompassing the areas of Cloud Computing, Data Mining, Image Processing, and Wireless Sensor Networks, Big Data, Semantic Web, Software Defined Networking and Fuzzy Automata. The activities are carried out to provide technological IT solutions for the societal needs. The School has the well-established infrastructure and tools that are necessary for research. So, of the ongoing research activities are:

1. Resource Allocation in Cloud

best engineering college in india

Complexity of resource scheduling has increased as cloud computing is shifting from homogenous to heterogeneous nature, but still many aspects are unknown as it is in experimental stage.

The important aspects of Cloud Computing are:


In the era of traditional infrastructure management, cloud implementation was seen as another IT cost item, with on- or off-premises consumption being paid for in advance on contract. In contrast, today's cloud services have opened an opportunity to pay for IT consumption. Designing and optimizing an IT environment that responds to demand in real time is one of the biggest challenges facing cloud users given the spikes and troughs in typical requirements.

Load balancing is essential for optimization of resources in distributed environments. The major goal of the cloud computing service providers is to use cloud computing resources efficiently to enhance the overall performance. Load balancing in cloud computing environment is a methodology to distribute workload across multiple computers to achieve optimal resource utilization with minimum response time. Although cloud computing has been widely adopted, research in cloud computing is still in its early stages, and some scientific challenges remain unsolved by the scientific community, particularly load balancing challenges:


The cloud service providers provides the services based on the service level agreement (SLA) established with the client, so in order to maintain the SLA and to ensure high availability of the service, the cloud datacenter deploys very high performance servers and other infrastructures. These datacenters are very much expensive to maintain as they consumes very large amount of electricity and additional power is consumed in cooling the server and other computational devices. Following are the various statistics related to the energy consumed by the datacenters:


Besides consuming very large amount of electricity and generating the huge bill for the cloud provider, these datacenter also generates very large amount of the greenhouse gases like carbon dioxide (CO2). Due these mentioned reasons there has been huge pressure from the governments all over the world green –activist to make the cloud datacenter eco-friendly.

The main reasons for the high energy consumption is not just the large number of power inefficient hardware devices but the various algorithms and policy the cloud service provider deploys to harness the computing power of these devices.

The physical computing servers at the cloud datacenters are never idle, most of the time they operate at 10-15% of their full capacity and their utilization rarely reaches 100%. The problem with this utilization model is that even when the servers are in idle state they still consume 70% of their peak power. This energy inefficiency of the datacenters can be prevented by leveraging the virtualization technology which creates multiple virtual machines on the single physical server thus by maximizing the utilization of the resources. It can also be prevented by using the live virtual machine migration which consolidates the VMs to the less number of servers depending upon their current resource usage pattern and switching the remaining idle servers to the less power consuming modes such as sleep or hibernate. The major problem with the live VM migration is that the resource usage pattern of the modern service application can’t be predicted and hence frequent live migration of virtual machines can lead to performance degradation and the service level agreement signed between the consumer and cloud provider can be violated. The above-mentioned problems have clearly stated the need of efficient resource scheduling which can efficiently allocate the virtual machines to the available servers and execute the users request i.e. task to the appropriate virtual machine, by maintaining the energy-performance trade-off in the cloud computing system.

2. Wireless Sensor Networks

best engineering college in india

Abstract

Wireless Sensor Network (WSN) consists of energy-constrained and storage-limited nodes equipped with sensor, processor, memory and wireless communication device able to communicate among them using wireless interfaces and collaborate to perform automated tasks requiring sensing capabilities. Routing protocols for wireless sensor networks are used to transmit messages from source to destination. They can be classified as unicast, broadcast and multicast. There have been a lot of multicast routing proposals for Ad-hoc networks, each of them based on different design constraints and decisions, application specific. Unfortunately, they cannot fulfil the unique requirements of WSNs effectively.

Objectives

3. Rule Based Resource Management (RBRM) in Cloud Environment

best engineering college in india

Introduction:

The specialty of virtual desktop cloud computing is that user applications are executed in virtual desktops on remote servers. Resource management and resource utilization are very much significant in the area of virtual desktop cloud computing. Handling a large amount of clients in most efficient manner is the main challenge in this field. This is because we have to ensure maximum resource utilization along with user data confidentiality, customer satisfaction, scalability, minimum SLA violation etc. Assigning too many users to one server may cause overloaded condition and which will lead to customer dissatisfaction. Assigning too little load will result in high investment cost. So we have taken in to consideration these two situations also.

Research area:

Rule Based Resource Management (RBRM) scheme assures the above mentioned parameters like minimum SLA violation. The concept of virtual desktop cloud computing is extended to a hybrid cloud environment. This is because to make the private cloud scalable. And priorities are assigned to user requests in order to maintain their confidentiality. The results indicate that by applying this RBRM scheme to the already existing overbooking mechanism will improve the performance of the system with significant reduction in SLA violation.

Approach (with block diagram)

We assume that there are M hosts in the data center and N users are subscribed to the service in the hybrid environment. All hosts are having the limited processing power, which is modelled as FLOPS (Floating Point Operations per Second). Based on the FLOPS requirements the resources are allotted which will be less than their worst case requirement. This reservation of resources in advance can be named as Resource International Journal of Computer Applications (0975 – 8887) Volume 66– No.7, March 2013 24 overbooking which will be discussed in the following section. The overall system model is given in figure-1:

Here a Rule-Based resource manager is implemented which will monitor the priority of the incoming requests. If the priority is high it will certainly be allotted to the private cloud. If the priority is low then it will be forwarded to the public cloud. The resource allocation and reallocation based on the “3-Rules” are discussed in the following sections. And here one more level of SLA optimization is introduced

Experimental results

The main functions that are implemented under the rule based resource management scheme are as follows:


best engineering college in india

Figure-2: Performance comparison of RBRM (priority based) approach with the cost-based strategy

Conclusion

The concept of virtual desktop cloud computing, i.e. executing applications in virtual desktops on remote servers, is very interesting because it enables access to any kind of application from any device. Here the optimization of the quality experienced by the customers by optimizing the distribution of the customers among the available hosts is carried out. In order to improve the quality experienced by the users the number of SLA violations experienced is reduced. And also implemented the minimum migration policy for each of the virtual desktops.

First, an optimization has been introduced to increase the average utilization on a single host. It was shown that the proposed overbooking approach, together with an advanced scheduler. To further optimize the quality of the service, a reallocation algorithm has been proposed to rebalance the virtual desktops among the available hosts after a busy period. After a busy period, some hosts could still be fully loaded while other hosts are almost not loaded and therefore, reallocating virtual desktops from fully loaded hosts to not loaded hosts can minimize the probability on SLA violations. The reallocation is done with the help of minimum migration policy for each of the virtual desktop. The enhancement can be made to the system by extending the concepts to the hybrid cloud environment and also can introduce the concept of “Rule-Based Resource Manager”. It will help to migrate the requests according to their priority and also is highly scalable and provides high resource utilization with minimum SLA violation.

4. Conversion of Textual-Visual Brain EEG Activities into Speech

Abstract

This work aims at developing system which converts visual brain signals related to text into speech. This can be further used in Brain Computer Interface (BCI) system in which mental tasks related to visual text data can be converted into voice-based commands. In this, mental activities when subject is visualizing the textual data are extracted through EEG and convert them into speech. This system also helps subjects with speech impairment in communication.

best engineering college in india

Methodology

A. Brain Signal Acquisition- EEG headset is used to capture signals corresponding to various mental tasks. Subject is asked to wear EEG headset which is non-invasive. Subject(s) are instructed to do any one from the following tasks during which signals are acquired.

Baseline Task – In this, subjects should relax. This state of relaxation should be retained without any motor actions.

Words Reading Task – The subject will either read the given words or sentences during which raw signals are captured.

best engineering college in indiaB. Pre-processing of Signal -The captured signals are raw signals. It contains lots of noises and artefacts. There will be many signals generated related to each mental tasks happening. Frequency of these signals will be very low. Pre-processing of these signals forms a very important step. This avoids errors and misinterpretations during classification. An Efficient filtering techniques like wavelet transform, signal filtering, amplification will be used. After removing spikes and other artefacts, signal will be ready for classification. Multiplexing- This is used for transmitting signal through BCI system. Also to combine the signals and make them to be in uniform distribution range so that it will be helpful for classification. Signal Analysis Signals- are analyzed to know their features and characteristics which are helpful for classification and further implementation. E. Data Bank Training data sets will be stored in this. This dataset is used by classifier to map input signal features with prestored features which helps to take decision about which signals component belongs to which particular class. Training dataset holds the aprior pattern of signal for corresponding EEG range signals. best engineering college in india

C. Training and Classification: This step consists of training the classifier with many samples of data collected by acquiring mental tasks of various subjects.

D. Classification: Associative rules will be used in combination with Bayesian classifier. Association rules describe association relationships among the attributes in the set of relevant data. The rules will be of the following form: Body ==> Consequent [Support, Confidence] Body: represents the examined data, Consequent: represents a discovered property for the examined data. Support: represents the percentage of the records satisfying the body or the consequent. Confidence: represents the percentage of the records satisfying both the body and the consequent to those satisfying only the body. Association rules used in the proposed work will be of the form: Signal. Type = {frequency, amplitude}. For Example, signal. Class = Alpha [percentage of occurrence of frequency between 8-13, percentage of occurrence of amplitude range 2-100] Here classifier will Find all possible sets of signal items (signal sets) that have support (number of signals) greater than the minimum support (large signal sets). Make use of large signal sets to generate the desired rules.

Classification is achieved by applying associative rules in two levels:

Layer 1: rules are applied to classify raw incoming signal in to different bands of EEG signal like alpha, beta, gamma and delta. Each of these classes of signals corresponds to particular cognitive tasks.

Layer 2: classification task-specific signals from the above classes like SP-speech. Where, SP class signals are converted into audible range and announced as speech through speaker. All other classes of signals are considered as noise.

5. Syntax and Semantic parsing of Kannada Text with an insight to Word Sense Disambiguation

Introduction

Natural Language Processing (NLP) is the branch of computer science focusing on developing systems that allow computers to communicate with people using everyday language. NLP must require Computational Linguistics which concerns with how computational methods can aid the understanding of the human language

best engineering college in india

There are many phases of natural language processing technique, where each phase has to undergo complex procedures. Finally the output will be the complete understanding and meaning of the natural language sentence as shown in Figure 1.

Syntactic interpretation (parsing)

Syntactic parsing is the task of recognizing a sentence and assigning a syntactic structure to it. This stage includes finding of the correct parse tree showing the phrase structure of the string. Syntax concerns with the proper ordering of words, its structure and its effect on meaning.

Semantic Interpretation

At this stage, the extraction of the (literal) meaning of the string or sentence is to be carried out. Semantic analysis is the process of understanding the meaning of linguistic input (construction of meaning representations). It processes the language to produce common-sense knowledge about the world (extraction of data and construct models of the world). Semantics concerns the (literal) meaning of words, phrases, and sentences. The two semantics considered are Lexical semantics and Compositional semantics. The lexical semantics include meanings of component words and word sense disambiguation. Compositional semantics involves how words combine to form larger meanings and finally semantic analysis is the total understanding of the language.

Pragmatic Interpretation: At this stage the effect of the overall context on altering the literal meaning of a sentence is considered. Pragmatics concerns with the overall communication and social context and its effect on interpretation.

Natural Language Processing Architecture

best engineering college in india

KWIC: Key-Word-In-Context, HMM: Hidden Markov Models, MT: Machine Translation, MAT: Machine Aided Translation, DTP: Desk Top Publishing, OCR: Optical Character Recognition

Figure 2 is a summary of all the major tasks in language processing. Corpus is the basis for all aspects of language processing. Hence it should created by carefully collecting representative texts in electronic form. A corpus can be created by three main methods - i) entering the texts through the keyboard using an editor, word processor or a Desk Top Publishing (DTP) software, ii) by scanning the printed or hand written texts using a scanner and then using an Optical Character Recognition (OCR) software to transform the scanned image into an editable text, and iii) by reading in texts into a microphone and using a Speech-to-Text software to convert the read sounds into editable text.

A careful study of these words required for the selection of entries build a Lexicon or a dictionary for a specified purpose. Using a Key-Word-In-Context (KWIC) tool, all the sentences in the corpus in which a given key word occurs are listed down that helps to study the different senses in which the particular word is used in different contexts.

Morphology is an essential component in any language processing task. It is the study of internal structure of words and plays very important role in analyzing free word order languages. Indian Languages and even more in the case of Dravidian languages like Kannada exhibit a very rich and complex system of morphology. A morphological analyzer takes a complete word form as input and produces an analysis of its structure in terms of the constituent morphemes.

A dictionary contains all possible Parts of Speech (POS) for a word, irrespective of various contexts in which that word can be used. A statistical POS tagger is a software that is based on statistical techniques like Hidden Markov Models (HMM) which tags words in a text with the correct POS and other grammatical features considering the context in which the word is used in that sentence.

A computational grammar needs to be in more detailed and precise form than the traditional grammars which are meant for people as syntactic parsing depends on it. A parser helps to create a parsed corpus which in turn can be used for context based spell checking, tagging, OCR, speech recognition, grammar checker etc. A Machine Aided Translation (MAT) system can be built using bilingual lexicons, parsers, morphological analyzers and generators etc. A parallel corpus in two languages is very useful for building MAT systems.

Problem Identification

In the current work, it is proposed to implement syntax representation and semantic parsing of Kannada language. The issues of Word sense disambiguation of the language will also be addressed.

In syntactic parsing, ambiguity is a particularly difficult problem since the most plausible analysis has to be chosen from an exponentially large number of alternative analyses. From tagging words to full parsing of sentences, algorithms have to be carefully chosen that can handle such ambiguity. Writing down a context free grammar for the syntactic analysis of natural language is problematic since unlike a programming language, natural language is too complex to simply list all the syntactic rules in terms of a context free grammar. Moreover syntactic structures just see the grammar rules of any particular language, but many grammatically correct sentences have no meaning. For example: “A rat ate a cat” which is meaningless. Syntactic rules do not consider the interactions between different components in the grammar. Hence an efficient semantic analysis would also be made to infer the meaning of a particular sentence. Previous works on Kannada text have been done only for simple sentences; hence all types of Kannada sentences would be handled.

Applications of Natural Language Processing are many like machine translation, text processing, information retrieval, speech recognition and so on. Hence all the stages of Natural Language Processing on Kannada need improvement to get the benefits of NLP applications for Kannada people. It can be seen that the work already done or being done in Kannada is very small compared to what all needs to be done.

Objectives

Methodology


Possible Outcome

Given an input sentence, a parser produces syntactic analysis of that sentence. Kannada sentence is taken as an input text, which is added to syntax analysis and next applying the semantic role on output provided by syntactic parser. WSD that is depending on these two parsing techniques would also be addressed.

Patents

Sl. No. Name Title Year
1 Dr. Pritham Gajkumar Shah and Rajat Pugaliya An apparatus and method based for smart shopping by use of association rule of Data Mining. 2017
2 Dr. Pritham Gajkumar Shah and Dr.Hariprasad S A An apparatus and method based on single coordinate public key with one extra bit by using elliptical curve cryptography for the resource constrained wireless sensor network. 2017
3 Dr. Pritham Gajkumar Shah and Rachith Gajwani An apparatus and method based on smart filter for eliminating Smog and pollutants in the air attachable to automobile. 2017

Ongoing Funded Projects

Sl. No. Project Title Name of the Guide Funding Agency
  1. Smart Water Impurity Detection System Prof. Manjunath C R KSCST - IISC