Nitish

RESEARCH, TEACHING, or OTHER INTERESTS

Artificial Intelligence, Computer Science, Computer Vision and Pattern Recognition, Human-Computer Interaction

Scopus Publications

Robust Watermarking of Loop Unrolled Convolution Layer IP Design for CNN using 4-variable Encoded Register Allocation
Anirban Sengupta, Vishal Chourasia, Aditya Anshul, and Nitish Kumar
IEEE
This paper presents a novel hardware security methodology that uses register allocation phase of high level synthesis (HLS) process, 4-variable encoded signature, and loop unrolling for securing convolutional layer intellectual property (IP) design for convolutional neural network (CNN) (used in consumer electronics (CE) systems). The proposed method demonstrates improved security in terms of resistance against tampering and probability of coincidence compared to previous approaches, at nominal design cost overhead.


M-HLS: Malevolent High-Level Synthesis for Watermarked Hardware IPs
Anirban Sengupta, Aditya Anshul, Vishal Chourasia, and Nitish Kumar
Institute of Electrical and Electronics Engineers (IEEE)
Hardware Trojan insertion in high-level synthesis (HLS) generated intellectual property (IP) designs can pose strong security concern for the designers. Backdoor hardware Trojans can be inserted in the HLS design flow to compromise the produced register transfer level (RTL) IP design. This letter presents a novel malevolent HLS (M-HLS) framework introducing the possibility of two different hardware Trojan insertion [i.e., performance degradation hardware Trojan (PD-HT) and Denial of Service hardware Trojan (DoS-HT)] in multiplexer (mux)-based interconnect stage of HLS generated watermarked IP design. The proposed framework is validated on the watermarked MESA Horner Bezier’s IP, which indicates strong performance degradation and DoS achievable by an attacker at minimal area and power overhead.


Digital Fencing - A Solution to Animal-Human Conflict
Rohit Beniwal, Parshant, Nitish Kumar, and Niteesh Rathore
IEEE
Fencing of animals is vital because of several factors such as providing safety and protection to them, containing them, preventing the spread of illness, and avoiding animal-human conflict. However, physical fencing of animals has various challenges such as being time-consuming, expensive, unmaintainable, and in some cases ineffective also. Therefore, in this research paper, we propose a digital fencing mechanism using a Deep Learning algorithm to tackle the abovementioned challenges. To be particular, we design a ready-to-use version of yoloV5, which aims at providing a reliable, secure, and easy-to-implement solution for the animal-human conflict. Moreover, we used an object detection technique for the classification of the animals that come in a certain region covered by the cameras. Thus, our approach does not affect animal rights as they are not being continuously tracked like the GPS and other methods of prevention unless they appear in a restricted area. Hence, this concept of digital fencing can do miraculous work in this field as it can help farmers protect their farms without investing too much effort and time and by also keeping the required safety measures for the animals as well as for themselves. To implement this work, we created our own dataset of 10,000 images belonging to seven different categories of animals. As a result, we achieved classification with a precision of 0.872, recall of 0.858, mAP@0.5 of 0.904, and mAP@0.5:0.95 of 0.584.


Advancing Pneumonia Detection Using Deep Learning
Nitish Kumar, T. Sai Keerthi, P. Varshini, P. Vijay Kumar, A. Harish, and R. Kumuda Vasini
IEEE
Deep learning has made significant advancements and transformative contributions to various aspects of the medical field. Its applications range from medical image analysis to diagnosis and prediction of diseases such as pneumonia. A type of infection known as pneumonia is brought on by bacteria, viruses, or fungus. It results in fluid or pus accumulation in the lungs' air sacs, called alveoli, which results in difficulty to breath. Early detection of pneumonia is crucial for avoiding adverse effects, including death. X-ray imaging is the most frequently used method for detecting pneumonia, although there are other methods as well, such as pulse oximetry and CT scans. The primary goal of this work is to create a precise model that aids in the detection of pneumonia and the category which it belongs to i.e., viral Pneumonia, bacterial Pneumonia. As part of the diagnosis, determine pneumonia using chest X-ray pictures. In the proposed work, deep learning Models like Inception V3 and MobileNetV2 are used to leverage their deep architectures and powerful feature learning capabilities to efficiently On the basis of evaluation criteria like accuracy, precision, and recall on the dataset of chest X-ray images, the suggested model's efficacy is then assessed. The existing work for Pneumonia Detection is Convolution Neural Networks (CNN) which achieved an accuracy of 84.12% this accuracy can be improved by the proposed model.


Approaches towards Fake News Detection using Machine Learning and Deep Learning
Nitish Kumar and Nirmalya Kar
IEEE
Fake news evolving around us for a very long time. The gradual growth of social media platforms has provided us with an easily accessible and publishable news platform in front of the audience that news may be true or False. The spreading of fake news increased as compared to ancient times. Nowadays Fake news detection become a tough challenge for Both Natural language processing(NLP) and Machine Learning (ML) experts. For detecting fake news fact-checking is also very important. In this paper, we focus on the analysis of recently published papers in this domain and the analysis of different techniques for detecting fake news. Through this survey, we will get inside knowledge of the detection process of fake news using different natural language processing, machine learning, and Deep Learning Techniques.


Deception Detection using a Multimodal Stacked Bi-LSTM Model
Puneet Kumar Sehrawat, Rajat Kumar, Nitish Kumar, and Dinesh Kumar Vishwakarma
IEEE
In the scientific community, researchers have recently become interested in the automatic identification of deceptive actions because of the range of fields where it might be advantageous, including criminology or security. Deception detection in conversational speech has drawn much attention in recent years. Given the significant risks associated with trial outcomes, using precise and efficient computational methods to assess the accuracy of court evidence may be very beneficial throughout the decision-making process. This study discusses about spotting deception in trial data from actual cases. Doing this has some challenges associated with creating a robust model that can accurately classify the deception and perform this action as fast as possible. Due to the limited number of videos and datasets available, some models overfit the data. A model should exist which can classify the data using various modalities, i.e., video, audio and text, and be able to work on multiple different datasets with excellent accuracy. This study has used videos from actual trials that were collected from open court proceedings and some videos from other datasets. To design a robust deception detection system that discriminates between witnesses and defendants, genuine and fraudulent testimony, this study investigates the utilization of text, audio and video modalities. By extracting and integrating information about the spoken words from audio, this study can achieve an accuracy of 80% approximately. The proposed model results with a classification accuracy of 96% approximately in an extended approach to perform video transcriptions. The Bag-of-lies dataset, a multimodal database captured in real-world settings has achieved an accuracy of 85%. The Miami University Deception Detection Dataset focuses on people telling truths and lies about their social relationships, achieved an accuracy of 98.1 % on the presented model. The proposed model employs LSTM (Long-Short Term Memory), Bidirectional LSTM (Long-Short Term Memory), CNN (Convolution Neural Network), and RestNet50. The results demonstrate that the proposed algorithm performs better at detecting deception than humans.


Operating Systems Support and Network Optimization View of Internet of Things
Vikas Kumar Agarwal, Nitish Kumar, Rajesh Singh, Anuradha Pathak, and Bharat Bhushan
Springer Singapore



Countermeasures of Different Jamming Attacks in Wireless Sensor Networks
Vikash Kumar Agarwal, Amit Kumar Rai, and Nitish Kumar
Springer Singapore



A dynamic session oriented clustering approach for detecting intrusions in databases
Indu Singh, Poornima, and Nitish Kumar
Springer Singapore



Automatic test data generator: A tool based on search-based techniques
Ruchika Malhotra, Poornima, and Nitish Kumar
IEEE
Software Testing is the most time consuming activity in the software development lifecycle. It is impossible to test everything. Hence, several automated test data generation techniques have been introduced in recent times in order to reduce the effort spent during testing. Search based techniques have been found to be more efficient than normal or random testing. In this paper, we propose to demonstrate the designing framework, implementation and explore the capabilities of a tool to aid in the generation of test data. Our tool is based on generating the optimal set of test cases based on the user defined coverage criteria. We have implemented the system in C++ language and have restricted ourselves to the use of command line interface. We provide the path as well as the test cases generated to the tester making his work of testing a lot easier.