• DAMM: Dynamic Modality Agnostic Weighted Embeddings Fusion for Multimodal Meme Detection
  Authors: Mohsin Imam, Utathya Aich, Ram Sarkar
  Venue: Elsevier, 2024 (Under Preparation)
  [link] [abs]

  Abstract: Coming Soon!

• Applications of Multimodal Deep Learning in Medicine
  Authors: Pankaj Rajdeo, …, Mohsin Imam, VB Surya Prasath
  Venue: Springer Handbook (SHB) on Medical Biotechnology, 2024 (Under Review)
  [link] [abs]

  Abstract: Coming Soon!

• On Utilizing Deep Learning Models for Preventing Harmful Response in LLMs
  Authors: Mohsin Imam, Salehah Hamzah, Shalini Aggarwal, Soumyabrata Dev, V.B Surya Prasath
  Venue: Neural Networks, Elsevier, 2024 (Under Review)
  [pdf] [abs]

  Abstract: Past few years have witnessed tremendous progress in artificial intelligence (AI) technologies including but not limited to natural language processing (NLP), metaverse, and generative models. These changes mark a new era with large language models (LLMs) like generative pre-trained transformers such as the GPT-3, GPT-4, and Llama2 gaining importance. These models, trained on vast text data, generate human-like text and find use in content creation, Q & A systems, and chatbots. Machine learning advances enable models like ChatGPT to respond to real-time text prompts. However, they are vulnerable to attacks, leading to inaccurate or harmful content. Adversarial prompting addresses these issues, including input manipulation, creating inducive prompts which causes LLMs to produce illegal content, simulating violence and harm, privacy breaches and other illegitimate actions. LLMs are especially prone to text attacks, with minor changes resulting in different outputs. Researchers are improving model security through adversarial training. To enhance the security of LLMs and prevent them from responding to adversarial prompts, we aim to utilize machine learning, deep learning and transformer models as external prompt classifiers for LLM-based applications like ChatGPT, to detect and counter malicious prompts that may lead to inappropriate responses from LLMs. We have conducted analyses on multiple machine learning, deep learning, and Transformer-based models, incorporating techniques such as ensembling and attention mechanisms. These models are trained on word-level text classification to categorize prompts into the adversarial category, preventing them from being answered by LLMs. We compared our results with those of other ML and DL models, and our proposed ensemble of the BERT-DistilBERT model achieved the best outcome with 97.56% accuracy in identifying malicious/adversarial prompts.

• Enhancing BERT-Based Models for Detecting Race, Violence, and Hate Speech in Low-Resource Settings
  Authors: Salehah Hamzah, Mohsin Imam, Masnizah Mohammad
  Venue: Journal Name, 2024 (Under Review)
  [link] [abs]

  Abstract: Coming Soon!

• An Ensemble Method for Categorizing Cardiovascular Disease
  Authors: Mohsin Imam, Sufiyan Adam, Neetu Agarwal, Suyash Kumar, Anjana Gosain Venue: International Conference on Advances in IoT and Security with AI, 2023
  [link] [abs]

  Abstract: Recently, machine learning models have become a key methodology in detection of cardiovascular diseases (CVD). This gives medical practitioners diagnostic support and indicators. In this work, we compare various machine learning (ML) classification algorithms, apply them to disease dataset and examine how these algorithms perform when subjected to either of the classes to aid in the study and investigation of CVD through computer-aided diagnosis (CAD). Our two main goals in this work are to first offer an automated machine learning ensemble model for categorizing cardiovascular malignancies and second to compare the performance of several classification algorithms to find the best classifier for the task. The proposed technique is specifically developed as a potential support for clinical care based on patient diagnostic data. The proposed approach exhibits an accuracy of 94.28% in the detection of cardiac illnesses when a thorough examination of binary classification is performed and averaged over numerous model training iterations. We believe that incorporating the suggested ensemble methods would produce stable and dependable CAD systems.

• Study of Student Personality Trait on Spear-Phishing Susceptibility Behaviour
  Authors: Mohamad Alhaddad, Masnizah Mohd, Faizan Qamar, Mohsin Imam
  Venue: International Journal of Advanced Computer Science and Applications (IJACSA), 14(5), 2023
  [link] [abs]

  Abstract: Spear-phishing emails are an effective cyber-attack method due to the fact that the emails sent are highly personalized to look like a regular legitimate email. Recently, it was discovered that personality traits of the victim have an impact on a person's susceptibility to spear-phishing. This study aims to identify which personality traits affect spear-phishing susceptibility besides other traits such as Information Technology background, gender, and age. In addition, measure of the effectiveness of embedded training systems and see whether message framing can further help increase its effectiveness. A personality trait survey was sent to 100 participants, followed by a real-life spear-phishing simulation to measure a certain personality trait’s influence on phishing susceptibility. After a two-week period, the second round of spear-phishing emails was sent again to measure message framing effectiveness. The personality traits analysis results show that users with higher levels of Internet anxiety are less susceptible to spear-phishing emails. While the message framing did not show any significant results, the embedded training program reduced the click rate. Findings revealed that certain people are more susceptible to spear-phishing emails than others. Thus, this work can guide an institution or organizations to identify which group of people are more vulnerable to spear-phishing.

• Review on Applications and Security of Blockchain
  Authors: Mohsin Imam, Kavita Saini
  Venue: Book Chapter: Blockchain and EHR, Nova Science Publisher, 2023
  [link] [abs]

  Abstract: The evolution of industry has been significantly influenced by blockchain technology. Regarding security, data access, auditing, and transaction management inside digital platforms, blockchain decentralized technology and privacy protocols offer potential benefits to many businesses. Blockchain is based on distributed and secure decentralized protocols; there is no central authority or point of control, and the network's nodes themselves generate, add, and validate the data blocks. Finally, Blockchains enable transparency by allowing each participant to observe transactions at any moment. Smart contracts provide secure transactions, reducing the risk of third-party disruption. Ethereum is a decentralized platform that facilitates the execution of smart contracts. This allows developers to design markets that move funds in accordance with instructions issued years ago. Decentralization and immutability are the principal characteristics of blockchain. Faster transactions, transaction and validation in seconds, and so forth. The security and applications of blockchain technology with brief introduction is discussed in this study. We can draw conclusion from this chapter article that there are security concerns with blockchain technology. These security concerns ought to have an impact on transactions as well. Blockchain technology offers some protection against certain attacks and also provides solutions to these problems.

• Adversarial Learning of Security Risk in Cobots Driven Industry
  Authors: Mohsin Imam, Mohd Anas Wajid, Aasim Zafar, Mohammad Saif Wajid, Neeraj Baishwar, Sameer Awasthi
  Venue: Book chapter: Intersection of Machine Learning and Computational Social Sciences, CRC Press, 2024 (Accepted)
  [link] [abs]

  Abstract: Cobots, or collaborative robots, have significantly increased productivity and efficiency across a number of industries. Cobots exist alongside humans and interact with crucial systems, but this growing reliance on them also poses security issues. By recognising and thwarting potential security threats and attacks, adversarial learning, a branch of machine learning, has emerged as a promising method to improve the security of Cobots. This chapter explores the use of adversarial learning in Cobot-driven sectors with a focus on identifying and reducing security vulnerabilities. We investigate different adversarial learning methods, including Generative Adversarial Networks (GANs) and adversarial training, and their potential uses in Cobot security. This chapter also explores how adversarial learning is used in Cobot-driven sectors, with a focus on identifying and reducing security vulnerabilities. The chapter also examines computational cost and scalability as well as other drawbacks of adversarial learning in this setting. To solve these issues and establish a secure environment for Cobots in industrial settings, we also propose prospective future research trends and advancements. The goal of this chapter is to support the trustworthy and safe integration of Cobots., ensuring human safety and encouraging the continuous expansion of Cobot-driven enterprises by understanding and utilising adversarial learning to minimise security threats.

• Cyber Threat Analysis and Mitigation in Emerging Information Technology (IT) Trends
  Authors: Mohsin Imam, Mohd Anas Wajid, Bharat Bhushan, Alaa Ali Hameed, Akhtar Jamil Venue: International Conference on Emerging Trends and Applications in Artificial Intelligence, 2023
  [link] [abs]

  Abstract: For the information technology sector, cybersecurity is essential. One of the main issues in the modern world is sending information from one system to another without letting the information out. Online crimes, which are on the rise daily, are the first thing that comes to mind when we think about cyber security. Various governments and businesses are adopting a number of actions to stop these cybercrimes. A lot of individuals are still quite worried about cyber security after taking many safeguards. This study’s primary goal is to examine the difficulties that modern technology-based cyber security faces, especially in light of the rising acceptance of cutting-edge innovations like server less computing, blockchain, and artificial intelligence (AI). The aim of this paper is to give readers a good overview of the most recent cyber security trends, ethics, and strategies. This study focuses on the present state of cyber security and the steps that may be taken to address the rising dangers posed by modern technology through a thorough investigation of the existing literature and actual case studies.

• Parametric Optimization and Comparative Study of Machine Learning and Deep Learning Algorithms for Breast Cancer Diagnosis
  Authors: Parul Jain, Shalini Gupta, Sufiyan Adam, Mohsin Imam
  Venue: Breast Disease, IOS Press, 2024
  [link] [abs]

  Abstract: Breast Cancer is the leading form of cancer found in women and a major cause of increased mortality rates among them. However, manual diagnosis of the disease is time-consuming and often limited by the availability of screening systems. Thus, there is a pressing need for an automatic diagnosis system that can quickly detect cancer in its early stages. Data mining and machine learning techniques have emerged as valuable tools in developing such a system. In this study we investigated the performance of several machine learning models on the Wisconsin Breast Cancer (original) dataset with a particular emphasis on finding which models perform the best for breast cancer diagnosis. The study also explores the contrast between the proposed ANN methodology and conventional machine learning techniques. The comparison between the methods employed in the current study and those utilized in earlier research on the Wisconsin Breast Cancer dataset is also compared. The findings of this study are in line with those of previous studies which also highlighted the efficacy of SVM, Decision Tree, CART, ANN, and ELM ANN for breast cancer detection. Several classifiers achieved high accuracy, precision and F1 scores for benign and malignant tumours, respectively. It is also found that models with hyperparameter adjustment performed better than those without and boosting methods like as XGBoost, Adaboost, and Gradient Boost consistently performed well across benign and malignant tumours. The study emphasizes the significance of hyperparameter tuning and the efficacy of boosting algorithms in addressing the complexity and nonlinearity of data. Using the Wisconsin Breast Cancer (original) dataset, a detailed summary of the current status of research on breast cancer diagnosis is provided.

• Assessing AI Chatbot Responses in Promoting COVID-19 Vaccine Acceptance
  Authors: Mohsin Imam, Abdul Quershi
  Venue: Social Science Research Network, 2024
  [link] [abs]

  Abstract: The global impact of the COVID-19 pandemic has extended across various facets of human existence, leading to a significant loss of life and overwhelming medical resources. Furthermore, the global economy has suffered considerably due to widespread job losses and decline, resulting in severe financial disturbances. Multiple section of society have made varied contributions to curb the virus’s propagation and safeguard public health. Notably, approval has been directed towards medical researchers for their dedicated endeavours in crafting COVID-19 vaccines. Rigorous clinical testing has confirmed the remarkable efficacy of these vaccines in mitigating symptomatic COVID-19 occurrences. However, a notable portion of the global population has exhibited reluctance towards receiving COVID-19 vaccinations. This reluctance stems from misconceptions surrounding the vaccines, a trend that has been exacerbated through a confluence of factors, encompassing the unrestricted access to internet-based information and the influential effect wielded by prominent personalities and authoritative figures. In this study, we have conducted an evaluation of responses furnished by the ChatGPT system concerning inquiries pertinent to vaccine fallacies. The affirmative and helpful perspectives expressed by the chatbot could potentially assume a pivotal role in moulding individuals’ outlooks on vaccinations. Additionally, these responses have the capacity to effectively motivate users to adapt vaccination, effectively minimizing prevalent misconceptions.

• Comparative Analysis of Deep Learning Models for Alzheimer’s Disease Stage Classification Using Transfer Learning
  Authors: Archana Gahlaut, Mohsin Imam
  Venue: Part of this work Venue: IEEE ICCCIS, 2023
  [link] [Report] [abs]

  Abstract: In recent years, advancements in healthcare technology, such as the rise of machine learning (ML), deep learning (DL), and artificial intelligence (AI), have sparked widespread interest due to their potential to improve survival rates and enhance people's health. Alzheimer's disease (AD), the most common neurodegenerative and dementing illness, is a major concern, with the cost of caring for patients expected to rise dramatically. Hence, the need for a computer-aided system to detect AD early and accurately is becoming crucial. Deep learning algorithms have proven to be more effective than traditional ML methods, as demonstrated by studies using brain MRI scans and convolutional neural networks (CNNs) for the diagnosis of AD. In the field of biomedical sciences, from drug delivery systems to medical imaging, predictive modelling and pattern recognition using ML and DL techniques have become essential for gaining a deeper understanding of complex medical problems. In this study, we compared various deep learning models for classifying four dementia stages related to AD by using transfer learning. The aim of classifying this type of medical data is to develop a prediction model or system to identify the disease from normal subjects or determine the stage of the disease. Classifying clinical data such as Alzheimer's disease (AD) has long posed a challenge, particularly in terms of selecting the most effective features. Our study successfully utilized various pretrained models, including EfficientNetB7, VGG, DenseNet, and others, to classify a dataset based on OASIS I and II extracted from Kaggle into Alzheimer's patients and normal controls. Our results showed an accuracy of 97.88% on test data using the EfficientNetB7 model. This experiment shows that the shift and scale invariant features extracted by pretrained models on large datasets, followed by deep learning classification, using the approach of transfer learning, is the most powerful method for distinguishing clinical data from healthy data in MRI. This approach also provides a path for predicting more complex systems in future.

• Precision Location Keyword Detection Using Offline Speech Recognition Technique
  Authors: Mohsin Imam, Gaurav Gupta
  Venue: Preprints.org, 2023
  [link] [abs]

  Abstract: This study introduces an original comprehensive system centered on identifying specific terms that indicate a user's position, particularly the discrete values representing latitude and longitude. This system not only detects these terms but also retrieves the corresponding numerical data for accurate and efficient determination of locations. The importance of this study can be applied various fields, notably aiding offline operations of military personnel, who often lack internet access. In such scenarios, precise awareness of location is vital for strategic manoeuvres, rescue operations, and navigating unfamiliar landscapes. The system allows these personnel by allowing them to extract exact location coordinates from spoken terms, thereby enhancing their awareness even in challenging surroundings. Apart from its military utility, the project holds broader significance. Teams responding to emergencies, personnel involved in disaster management, and exploratory missions can all gain from this technology during disruptions in communication infrastructure. Furthermore, travelers, adventurers, and outdoor enthusiasts can utilize this system to accurately determine their positions in remote areas without relying on online maps. We used offline speech recognition techniques to precisely transcribe spoken terms, achieving an accuracy of over 91.3% and a word error rate of 4.2%. For sound recognition, the OpenAI Whisper model was used, and a conversion process from SpeechRecognition to AudioSegmentation was implemented, followed by transforming the audio into .wav format, we have also developed the interface of the app to use it efficiently using Streamlit. This was done to ensure seamless compatibility with the Whisper model and uninterrupted audio input. By training the system to identify specific linguistic linked to location, it achieves robust detection and extraction of relevant terms. This approach eliminates the necessity for constant internet connectivity, rendering it exceptionally useful in remote, offline, and resource-limited situations.

• Air Quality Monitoring Using Statistical Learning Models for Sustainable Environment
  Authors: Mohsin Imam, Sufiyan Adam, Soumyabrata Dev, Nashreen Nesa
  Venue: Intelligent Systems with Applications, Elsevier, 2024
  [link] [abs]

  Abstract: High levels of air pollutants pose significant health risks, increasing the chances of respiratory infections, lung cancer, and heart complications, particularly for those already susceptible to illness. Modern societal advancements have worsened air quality degradation, with daily activities such as transportation, industrial processes, and domestic operations releasing harmful contaminants. This study addresses the urgent need for air quality monitoring and forecasting, especially in developing nations like India, where machine learning-based prediction technologies play a crucial role in understanding environmental aspects. Our study focuses on analyzing and predicting air quality using data from two distinct areas in Kolkata—Victoria and Rabindra. Thorough pre-processing and data analysis have been conducted to identify essential features and detect undermining patterns in the data. Utilizing five classic machine learning algorithms, we predict the air quality category by categorizing the predicted AQI into six AQI classes, emphasizing extensive hyper-parameter tuning for each model. The Rabindra dataset yields the best-performing Support Vector Machine (SVC) model with a 97.98% accuracy, while the best prediction accuracy for Victoria location is 93.29% from Random Forest Classifier (RFC). This study offers valuable insights into the effectiveness of machine learning algorithms for air quality prediction, with the novelty of the study lying in the focus on hyper-parameter tuning to achieve the highest accuracy.

• Building Blockchain-powered Decentralized Applications with Solidity: A Project-Oriented Guide
  Authors: Sparsh Sharma, Mohsin Imam
  Venue: To be Announced, Currently Under Process
  [Details]

  • Comprehensive Guide: "Building Blockchain-powered Decentralized Applications with Solidity: A Project-Oriented Guide" offers a detailed introduction to blockchain technology and DApp development, divided into three focused sections.
  • Fundamentals of Blockchain: The first section, "Introduction, Installation and Solidity Programming Language," covers blockchain fundamentals, DApp evolution, development requirements, and an introduction to Solidity programming.
  • Hands-on Projects: The second section, "Practical Projects on Blockchain," provides real-life project-based learning, such as developing a decentralized voting application, offering interactive and practical insights.
  • Advanced Integrations: The third section, "Blockchain with Other Advanced Technologies," explores blockchain's integration with technologies like IPFS, IoT, DAOs, DeFi, and NFTs.