Nurturing Undergraduate Research through the R&D Project
The R&D project was devised to enable NU students to work in the research being done by the faculty as co-investigators. Faculty share their research problems and the students are given an opportunity to choose the problem that interests them most. After which, they work with the faculty over the course of a semester on the research problem. They get to do literature survey, devising instruments to collect data, data gathering & analysis and finally presenting their research findings to a faculty panel.
NU students can also take advantage of the research assistantship programme – NURap (NU Research Assistantship Programme) and the travel assistance programme – NUTap (NU Travel Assistantship Programme) during the R&D project and after the project if they choose to continue doing research.
NURap provides a monthly monetary assistantship and can be availed of during the R&D project and afterwards too if the project work continues. NU students who co-author papers with their faculty-mentor can make use of NUTap which provides monetary assistance to cover expenses towards travel and presenting the paper at the conference.
Some R&D Projects undertaken by NU Undergrads
The Development of an e-tutor for the Java Language
The goal is to create an interactive e-tutor for the Java language, with features such as a clickable index, easy navigability to a page of interest, and the ability to place book-marks as well as to jot marginal notes in the text of an e-book for Java Programming. The programming for these capabilities has been completed. Textual content including example programs, illustrative diagrams, quizzes, etc. is being added. The software is being created using the Java programming language.
Mentor: Prof Sudhir Kaicker
Intrusion Detection Using Data Mining & Clustering Techniques
Intrusion in networks and other related fields are rampant in today's world. It is essential to identify such attacks and develop schemes for future prevention of such threats. Data mining, clustering techniques and their analysis can help mitigate such issues. Students got a chance to develop intrusion detection and prevention tools which could be implemented on network log data and such areas. They worked on the NSL-KDD dataset to devise an Intrusion Detection System. They managed to reduce the number of features (attributes) in the dataset to overcome the curse of dimensionality, without any compromise of the performance parameters. The favourable performance criterion was False Negatives < False Positives; True Positives > False Positives; True Negatives > False Negatives; and False Negative % ≈ 0.The performance was based on the Accuracy (most important factor) of the system and the time taken to provide a correct classification. Results obtained by the students proved to be better than results published in Elsevier and Springer papers during 2014, 2015.
Mentor: Prof Trupil Gordhanbhai Limbasiya
Hand Gesture-based HCI using Raspberry-Pi
The objective of the project is to identify a set of specific hand gestures captured and processed using Raspberry-Pi (with the camera module) and perform certain operations in a computer based on the recognized gestures. Hand gesture recognition involves resource intensive computations. One of the objectives is to verify the feasibility of implementation of both static and dynamic gestures in a resource constrained device like the Raspberry-Pi which is a single board computer with limited processing capability and memory. The project opens up the opportunity for writing interesting applications based on human computer interaction using hand gestures. As a user interface, hand gesture is preferred over keyboard or mouse based interaction; interaction with a computer can be done at a distance without any wired communication. There are a number of issues to be resolved - e.g. camera view point change, illumination related problem caused due to change in ambient lighting condition, gesture end point detection for dynamic gestures, etc.
Mentor: Prof Soharab Hossain Shaikh
Development of new potential antibacterial agents by using metal oxide nanoparticles in combination with antibiotics
Antibiotic resistance is a common problem in bacterial community and is a major challenge for scientific community to develop new antibacterial agents. One of the promising approaches to overcome antibiotic resistance is use of nanoparticles. Nanoparticles have antibacterial properties due to their small size, large surface area and their ability to target different biomolecules. Although, these nanoparticles have great antibacterial properties, they have also been reported to cause inflammation and tumour in mammalian cells. Combination of nanoparticles and antibiotics can be a better approach because fewer doses of both require which ultimately reduce the toxicity towards mammal cells. Moreover, nanoparticles coated with antibiotics have less interaction with mammal cells. Preliminary work done under this project showed that combined used of ampicillin and titanium dioxide (TiO2) nanoparticles effectively killed the ampicillin resistant E. Coli strains.
Alphonse Dhas Antony
Mentor: Prof Mandeep Dadlwal
Recommendations Using Distant Neighbors
Recommendation Systems have been successfully used to provide items of interest to the user. Typical Recommendation Systems that use location assume Tobler's First Law of Geography "Everything is related to everything else but near things are more related than distant things." The goal of this project is to study the influence of distant neighbors and incorporate the same in a recommendation algorithm. However in today's world, the notion of proximity has changed. The theory of "six degrees of separation" proposed by Hungarian author Frigyes Karinthy in 1929 has to be modified in today's connected world. Facebook reported in 2011 that amongst its 721 million users, the average distance was only 4.74. This further reduced to 3.57 amongst 1.6 billion users in February 2016. In this project, we try to partition the users into smaller communities based on proximity. The partitioning is done based on both geographic as well as social proximity, with a broader interpretation of Tobbler's law. This makes the recommendations more focused while allowing the algorithm to be run on smaller partitions of the data. This allows us to solve larger problem instances improving the scalability of the proposed solution.
Anurag Kumar Singh
Devjeet Singh Choudhary
Mentor: Prof Prosenjit Gupta
Improving Top-N Recommendation Techniques Using Rating Variance
The concept of this project involves the use of rating variance in the prediction of top N neighbours as a proxy for the consumer’s confidence in a given choice. A negative (left-side) skewness indicates that the general mass of the ratings lies near the upper end and that there is a long tail of lower ratings. Items with lower skewness in rating will therefore have higher means compared to the data set. Hypothetically, this is desirable. Therefore, an approach is tested involving the usage of low skewness. We reject those items with skewness > S, where S is some pre-defined constant. We give items with skewness > T a bonus to their rating, where T is some pre-defined threshold. In a combined approach, we use a higher threshold for filtering and a lower one for ranking. Kurtosis is a measure of the “tailedness” of the distribution. High kurtosis tends towards values being clustered in a single area. This is similar thematically to the idea of using lower variance in our recommender systems, and therefore we will also be testing the desirability of high kurtosis in ranking and filtering. Here we reject those items with excess kurtosis < K and allow a bonus to ranking for low excess kurtosis. In this study, using an adjusted ranking/filtering approach we have demonstrated that the accuracy of recommendations can be improved by filtering out recommendations above a minimum rating skewness threshold and giving those recommendations with particularly low skewness a boost in ranking. We conducted experiments with the well-known Movielens (20M) and Jester datasets. The Jester data set shows a higher affinity for skewness-related schemes as compared to the MovieLens 20M data set, possibly due to the large number of ratings for a given item.
Mentor: Prof Prosenjit Gupta
Geospatial Analysis of Sahibi River
The Sahibi River is an ephemeral, rain-fed river in India. It is also known as Sabi Nadi, It is the river which flows in the north direction from the Aravallis. The river is dried up from last two decades. The objective of this investigation will be two folds. The first is change in Sahibi river pattern and climatic parameters and second is to assess the implications of its revival. The area of river channel has been decreased from 85.85sqkm to 36.46sqkm, from 1972-2015. From LU/LC change analysis it has been clearly seen that the area of the water bodies had been reduced enormously from the year 1972 to 2015. In the groundwater level analysis it has been clearly seen that the groundwater level has been keep on decreasing which indicates that consumption is higher than recharge. The 2016 monsoon was good and water has been observed at some locations in Sahibi area. The work is in progress to assimilate the 2016 Remote Sensing and meteorological data into current investigation. The ways of revival could be draining the two canals (one from Narnaul and other near to Kotkasim) in the riverbed.
Mentor: Dr Mohd Anul Haq
Role of Body Area Sensor Networks in Smart Health Care
The term ‘computer’ or ‘PC’ has never yielded another term ‘personal’. It is just regarded as a machine with inputs from human beings giving required outputs. However the term or the budding field of Body Area Networks gives new meaning or leverage now to the term ‘personal’ in PCs. In short, this wireless technology leverages wireless communications protocols allowing low-powered sensors to communicate with one another and transmit data to a local base station and to remote places like hospitals. This paper on Wireless Body Area Networks will present a detailed discussion on the various applications of WBANs in smart healthcare systems, its history of development, advantages, disadvantages and the future prospect and scenario of this system. We will also highlight project performed earlier that helped WBANs to provide long-term healthcare monitoring. Our main aim in this project, is to understand one of the major applications of Body Area Sensor Networks, termed as ‘Amplifier Ear’ which helps people suffering from auditory problems to function properly.
Neelanjana Basu Roy
Mentor: Dr. Debasis Das
Algorithm for Secure Cloud Computing
Cloud Computing is a rising field of Information Technology (IT) in the history of computing. It is a way to maximize the capacity and capabilities without spending a lot to buy a new infrastructure and software. When users are online, they can get faster access to their data due to the massive storage. Although Cloud computing has many advantages due to a large number of organizations moving towards it, it comes up with lots of security issues and breaches faced by both cloud service providers and users which are addressed in this paper. An efficient framework is devised for dealing with such issues. Proposed framework can protect data while transferring, sharing and storing in data centers using the classification of data, Hashed Message Authentication codes and Index Building. The data are divided into three sections and accordingly the user is asked for authentication. The user is provided the digital signature which can be verified with cloud directory. Using indexing, search can be made on the encrypted data.
Mentor: Dr. Debasis Das
Improvised broadcast algorithm for wireless networks
Broadcasting problem is an important issue in the wireless networks, especially in dynamic wireless networks. In dynamic wireless networks the node density and mobility is high, due to several problems which arise during broadcasting. Two major problems faced are namely, Broadcast Storm Problem and Disconnected network problem. In a highly dense network, if information is being flooded in a loop, it could lead to broadcast storm. The broadcast storm may eventually crash the entire network and lead to loss of information. Mobility of the nodes may lead to the problem of Disconnected Network. If the two nodes sending and receiving information are mobile with different speeds, it could lead to a disconnection between them as soon as the receiver moves out of the communication range. In this project, we are trying to solve both the problems based on our proposed algorithms.
Mentor: Dr. Debasis Das
Algorithm for Leader Node Selection in Vehicular Adhoc Networks
With advancement in the field of vehicle automation and wireless communication, Vehicular Ad Hoc network (VANET) based solutions are emerging as the answer to the dire problem of road traffic management. As dedicated short-range communications (DSRC) devices, vehicles can be organized in a peer-to-peer network to manage their movement for a smoother traffic flow. Our paper targets traffic management at road intersections. Our approach calls for the selection of a leader node in every lane so that they may mutually decide a safe and efficient order of crossing the intersection. Generally leader node selection algorithms have a computational complexity of O(nlog(n)). Our proposed algorithm uses geographic routing coupled with the sequential flow of communication to select the leader node with a computational complexity of O(n) where n is the number of vehicles present in a lane at Road Intersection.