Computer Engineering Commons^™

Community Detection Using Deep Learning: Variational Graph Autoencoder Enhanced With Leiden And K-Truss Techniques, Jyotika Hariom Patil

Master's Projects

Community detection in networks is essential for understanding the complex structures of connected systems. Traditional deep learning (DL) methods such as Graph Neural Networks (GNNs) and Graph Convolutional Networks (GCNs) have shown promised results in supervised tasks, like classification, but often fail in unsupervised tasks like community detection because of the lack of labels. Self- supervised approaches where we integrate crucial community information offer a solution. This project seeks to explore DL methods for community detection, focusing specifically on using Graph Variational Autoencoders (VGAEs). While classical approaches can efficiently handle small to medium-sized networks, they typically struggle with larger-sized structures. …

Opinion Graphs Construction For Reviews Using Transfer Learning And Large Language Models, Yichen Lin

Master's Projects

With the rapid development of the Internet, reading online reviews before making a purchase, booking a hotel, or making a restaurant reservation has become a part of daily life. Customers often consider reviews as crucial supplementary information before making decisions on how to spend their money. However, reading many reviews to gain helpful information takes time and effort. This project proposes a new method OpinionGraphGenerator that aims to create opinion graphs from hotel reviews to reduce the high volume of text in reviews while preserving essential insights. In an opinion graph, vertices are semantically similar opinions, where each opinion consists …

Skin Cancer Detection Using Reinforcement Learning, Vikas Chercadu

Master's Projects

Advances in medical diagnostics have increasingly harnessed the power of artificial

intelligence, offering substantial improvements in early and accurate disease identifi- cation. This paper elaborates on a novel integration of Multi-Agent Reinforcement

Learning (MARL) with Deep Learning for the early detection of skin cancer, one of the most prevalent and lethal forms of cancer when left unchecked. Our project capitalizes on the sophisticated VGG16 Network for the extraction of detailed features from the widely-utilized HAM10000 dermatoscopic dataset, enhancing these features with additional color, texture, and shape analysis. Utilizing a custom-designed MARL environment, we facilitate a collaboration among multiple intelligent agents, …

Robustness Of Learning Models To Label Flipping Attacks, Sarvagya Bhargava

Master's Projects

In this paper we compare traditional machine learning and deep learning models

trained on a malware dataset when subjected to adversarial attack based on label- flipping. Specifically, we investigate the robustness of different models when faced

with varying percentages of misleading labels, assessing their ability to maintain their accuracy in the face of such adversarial manipulations of the training data. This research aims to provide insights into which models are more robust, in the sense of being better able to resist intentional disruptions to the training data. We find that traditional machine learning models and boosting techniques are more robust …

Mitigating Learning Bias In Healthcare Datasets, Samyak Jagdish Kumbhalwar

Master's Projects

CVDs have been a major cause of deaths worldwide with WHO reporting 17.9 million deaths annually. Although there are advancements in the treatment of these diseases, most of the fatalities are a result of untimely diagnosis. Active research is going on to collect data points and risk factors related to these diseases, which can enable early diagnosis. Of the datasets available, many researchers have employed different ML models to predict/detect the prevalence of heart diseases. Many employed Tree based, regression models [3, 6]. Few also tried ensemble approaches [1, 2, 4]. These healthcare datasets are generally found to be imbalanced. …

Gesture Recognition Dynamics: Unveiling Video Patterns With Deep Learning, Nithish Reddy Agumamidi

Master's Projects

This paper, Gesture Recognition Dynamics: Revealing Video Patterns with Deep Learning, explores the combination of Long Short-Term Memory(LSTM) with Convolutional Neural Network(CNN) in the identification of convoluted human activities. The study assesses LSTM’s capability to capture temporal dependencies and CNN’s potential to apprehend and extract spatial characteristics to detect the gestures from UCF50. It further evaluates the architecture linkage of LSTM and CNN, which will improve the analytical capacity to interpret and validate dynamic gesture trends. The paper utilizes Mediapipe, an open-source framework created by Google specifically designed for extracting poses. The Mediapipe tool is well-designed to track important body …

Deciphering Speech Through Vision: A Deep Learning Lip Reading System, Srujith Rao Ambati

Master's Projects

Lip-reading, a ubiquitous field between computer vision and speech processing, focuses on identifying what spoken words a person generates depending on their uttering lip movements. This paper presents a streamlined lip-reading solution that employs machine learning and deep learning. First, Our work utilizes the Multi-Task Cascaded Convo- lutional Networks to detect facial “landmarks,” including the face and lips region, and the aligns the face. The aligned faces are segmented to get the lip images. Lip images are preprocessed using the Real-Enhanced Super Resolution Generative Adversarial Network to enhance image resolution to identify subtle lip movement in video images: a critical …

Gender Classification Through Postural Analysis: A Comparative Study Of 2d Images And 3d Reconstructions., Prathamesh Dixit

Master's Projects

In computer vision, gender classification has become a vital task having applications in human-computer interaction, healthcare, and surveillance. In this study, we look at a two-step approach based on human joint information for gender classification. In this research, we use convolutional neural networks (CNNs).

With Leeds Sports Pose (LSP) dataset, we use a C5 pre-trained model to map and extract joint information from 2D RGB images and after pre-processing and background removal, we use PiFUHD to transform these 2D images into 3D representations. Next, we train our models on RGB images and joint images for both 2D and 3D representations. …

Understanding Distracted Driving And Gaze Patterns In A Driving Simulator: Simple Vs. Complex Challenges, Rishi Prabhat Narayana Sannala

Master's Projects

Distracted driving has grown in criticality over the recent years, given the numerous distractions that drivers face today, and which have further been magnified by the proliferation of in-vehicle technologies and mobile devices. Such distractions can seriously compromise a driver's ability to be fully focused on the road and to carry out timely responses and informed decisions that are key in minimizing the risks of a crash and maximizing road safety. The general aim of the research is to observe how simple versus complex distractors affect driving performance and gaze patterns. The eyetracker used is the Tobii Pro Fusion, synchronized …

Intelligent Caching Using Continuous Machine Learning In Named Data Networking, Sai Sameer Yanamandra

Master's Projects

Our project focuses on improving Named Data Networking (NDN), an alternative network architecture to traditional IP networks, particularly in unstable conditions where connections frequently drop, and data movement is unpredictable. In NDN, data is cached at various routers across the network, enhancing accessibility even amidst unstable connections. A key challenge we address is determining the optimal level of data redundancy in unstable scenarios. We aim to balance the need for data availability with the risk of excessive data duplication. Our solution involves developing a novel data caching approach for the NDN’s content store based on continuous machine learning. This method …

An Empirical Analysis Of Adversarial Attacks In Federated Learning, Rohit Mapakshi

Master's Projects

In this paper, we experimentally analyze the susceptibility of selected Federated Learning (FL) systems to the presence of adversarial clients. We find that temporal attacks significantly affect model performance in FL, especially when the adversaries are active throughout and during the ending rounds of the FL process. Machine Learning models like Multinominal Logistic Regression, Support Vector Classifier (SVC), Neural Network models like Multilayer Perceptron (MLP), Convolution Neural Network (CNN), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM) and tree-based machine learning models like Random Forest and XGBoost were considered. These results highlight the effectiveness of temporal attacks and the need …

Employing Large Language Models And Retrieval Augmented Generation For Enhanced Predictive Flexibility In Cancer Mortality Prediction, Mridang Kejriwal

Master's Projects

Today, cancer is a major health risk to thousands of people, and there are over a two-hundred different types of cancer. Luckily, over the past several years, the outcomes and survival rates have increased, all thanks to machine learning, specifically Recurrent Neural Networks (RNN) and Long Short-Term memory (LSTM) networks. However, the current prognostic models don’t allow healthcare professionals to adapt the variables to mimic all the different features of every type of cancer, resulting in a model that works but is not as accurate as it could be. This study explores improving the accuracy and adaptability of the current …

A Rule-Based Hybrid Translation Model For Context-Aware Speech To Indian Sign Language Tasks, Dania Jaison

Master's Projects

In India, one of the most significant tools to communicate with the Deaf and Hard of Hearing (DHH) communities is Indian Sign Language(ISL). The issue of a scarcity of computational resources for ISL is even more pronounced when it comes to the translation of written or spoken English into ISL. This paper proposes a rule-based model for translation where the spoken english sentences are translated into ISL by focusing on the specific syntactic and grammatical differences between these two languages. ISL uses a simplified syntax unlike the nuanced sentence structures in English — we omit most of the auxiliary verbs …

Comparitive Analysis Of Time Series Forecasting Using Frequency Informed Dense Neural Networks And Lstm, Avinash Mangalore Suresh

Master's Projects

Time series forecasting influences our lives on a daily basis, being a versatile tool in various application areas like environmental studies, finance, medicine and much more. While there are many established statistical and deep learning approaches to model time series data, each implementation comes with their own set of drawbacks or areas of improvements. Most of the existing deep learning architectures and research have focused on modeling time series data in the time-domain exclusively. However training deep learning models in the time-domain has some drawbacks, mainly due to the inherent temporal dependence of each time-step on the time-steps before it, …

Multi-Platform Cyberbullying Detection Using Nlp And Machine Learning, Chinmayi Lokeshwar Hegde

Master's Projects

The issue of cyberbullying is growing due to the online anonymity and due to online platforms having less repercussions. This research proposes for proactive measures to detect and prevent such behavior before it reaches the victim. By using data from various social media platforms and employing machine learning techniques, this research proposes an innovative system aimed at identifying and thwarting cyberbullying incidents preemptively. While existing methods have primarily focused on prediction and detection of cyberbullying incidents, there remains a significant gap in research regarding prevention strategies. This project aims to address this gap by leveraging machine learning, natural language processing …

Temporal Dynamics In Diabetes Prediction: A Sensor-Driven Time-Series Exploration, Monica Meduri

Master's Projects

Diabetes is a lifelong illness that, if not detected or managed appropriately, turns into serious complications. Correct glucose forecasting is critical to ensuring timely interventions, thereby minimizing risks of hyperglycemia and hypoglycemia, and optimizing the management strategies of the disease. Classical machine learning models have been applied in the blood glucose forecasting problem for a long time, however, usage of transformer-based architectures is still scarce within the literature. Due to the self-attention mechanism, transformers can capture temporal relationships very effectively, which makes them suitable for time-series data. TFT is a novel framework proposed here to utilize time-series data from CGM …

Intelligent Caching In Named Data Networking, Deep Pradipbhai Shah

Master's Projects

Named Data Networking (NDN) has a built-in caching capability that is enabled with the help of its Content Store. Caching in NDN has several benefits, such as reducing overhead on the producer side, avoiding a single point of failure, and reducing network load. The primary caching policy of the NDN architecture is to leave copies everywhere. However, this scheme induces significant cache redundancy. Existing advanced cache techniques either periodically share the entire list of cached content at a node or make a caching decision without knowing the cached content at other nodes in the network. We propose an intelligent cache …

Multimodal Retrieval-Augmented Generation: Design And Application, Charul Rathore

Master's Projects

The rapid advancement in generative AI and large language models have forever revolutionized how we synthesize data. This project explores and experiments with the potential of a multimodal Retrieval-Augmented Generation (RAG) framework for processing text, tabular and image data. Starting with prompt engineering techniques, we address their limitations in dynamic and domain-specific real world applications by building a multimodal RAG pipeline and evaluating it against human-generated ground truth. The project culminates in BrightMind.ai, a full-stack educational platform featuring novel personalized AI companions for context-aware and adaptive response generation. Its innovative capabilities extend to music, video, and code generation, setting it …

Leveraging Large Language Models For Transforming Student Information Into Actionable Data, Sree Hari Karri

Master's Projects

Admission season places significant demands on university committees, necessitating the review of vast arrays of documents to assess students’ competence. This project advances the development of an automated system designed to streamline this process by evaluating application materials such as Letters of Recommendation (LoRs), Statements of Purpose (SoPs), and resumes. Utilizing a variety of advanced Natural Language Processing (NLP) techniques, the system compares the performance of several Large Language Model (LLM) approaches. It also experiments with different data handling strategies, including the use of vector stores versus traditional context-based processing, to optimize model efficiency and accuracy. Special attention is given …

Personalizing Image Generation From Prompts Using Generative Ai, Mit Ramesh Jain

Master's Projects

The rapid advancements in Generative AI, particularly Text-to-Image (T2I) models, have opened up new possibilities for personalized image generation. Finetuning large T2I models for specific downstream tasks is a key approach to achieving tailored outputs. In recent years, Parameter-Efficient Fine-Tuning (PEFT) techniques have gained significant attention as a cost-effective and efficient solution for fine-tuning large models. Initially developed for fine-tuning large language models (LLMs), PEFT techniques have been extensively studied and compared in the context of language tasks. However, regarding the T2I domain, there is a lack of similarly exhaustive and detailed literature on PEFT. This research project, in the …

Implicit Personality Detection From User Behaviour In Recommendation Systems, Uzma Zubair Shaikh

Master's Projects

Recommendation systems are an integral part of any business, and a crucial factor in determining their success as these systems help businesses in marketing their products to the right kind of audience. Conventional methods of building recommendation systems such as collaborative filtering and content-based recommendation, although effective, suffer from limitations such as cold start and the data sparsity problems. Moreover, these methods aim at finding similar products as user’s past interactions rather than personalizing the recommendations. The upsurge in use of social media, over-the-top content (OTT), and e-commerce platforms has made the task of personalizing recommendations imperative, leading to the …

Load Balancing For Cloud-Based Applications, Nitish Ranjan

Master's Projects

Effective load balancing is critical in ensuring optimal resource utilization, reducing latency, and improving the overall performance of distributed systems. This report commences with a comprehensive literature review on existing load-balancing algorithms, examining their methodologies, strengths, and limitations within various computing environments, including cloud computing, data centers, and network traffic management. Despite significant advancements in this field, the dynamic nature of distributed systems, coupled with the ever-increasing demand for efficient data processing, poses ongoing challenges. In response, this study proposes a novel load-balancing algorithm to address these contemporary challenges. The approach leverages dynamic and hybrid load balancing, distinguishing it from …

Knowledge Graph-Based Multiple-Choice Question Generation, Durga Muralidharan

Master's Projects

Knowledge-based tests are widely used to assess knowledge on a specific subject and have many applications in education and professional certifications. These tests usually consist of Multiple Choice Questions (MCQs), where a question with a few possible answers is given. Along with the correct answer, three or more incorrect answers are provided, which are called distractors. MCQs are a popular method for these tests because they are easy to grade. These tests can check different levels of comprehension ranging from beginners to advanced by creating distractors that may confuse unprepared test takers. This project proposes the Knowledge Graph Multiple Choice …

Code Quality Enhancement: Evaluating Ai Code Generation With Software Metrics, Sirisha Krishna Murthy

Master's Projects

With the advancements in the stream of AI in the recent time and the evolution of Generative AI, it is a given that there is a need to effectively integrate AI into daily tasks, including Coding. When talking about Generative AI, one important thing to consider is prompting, which is that way to talk to the AI. Depending on specific needs and tasks the way we need to prompt AI can vary. With rapid development in the field, there are a lot of new benchmarks that evaluate the AI coders on correctness, but to effectively adapt AI into actual coding …

Facial Expression Mood Classification Using Machine Learning, Tiantong Li

Master's Projects

Facial expression classification is a powerful tool for understanding human emotions, with applications spanning human-computer interaction, healthcare, and entertainment. By analyzing facial cues, systems can interpret emotional states and adapt their responses, creating more personalized and emotionally aware experiences. One emerging application of facial expression classification is in music recommendation systems, where user emotions are integrated to suggest music that aligns with their current mood. While prior research has primarily classified facial expressions into four emotion categories, this study broadens the scope to seven emotions: angry, disgust, fear, happy, neutral, sad, and surprise. The project evaluates four machine learning techniques—CNN, …

Cluster Analysis For Concept Drift Detection In Malware, Aniket Mishra

Master's Projects

The rapid evolution of malware presents significant challenges for detection systems. This is due to malware families adapting through feature manipulation and obfuscation, which causes concept drift. A clustering based approach is used to detect and adapt to these shifts. The KronoDroid dataset is segmented into batch sizes of 50 and analyzed with MiniBatch K-Means clustering. The silhouette coefficient is used to evaluate clustering quality, and help identify drift by detecting significant changes in cluster patterns. Concept drift will cause retraining of supervised classifiers, including Linear SVM, RF, MLP, and XGBoost. Three scenarios are used: static models, periodic retraining, and …

Detecting Crustose Coralline Algae (Cca) In Marine Photos Using Mask R-Cnn, Vrushali Harshwardhan Deshpande

Master's Projects

Coral reefs, made up of thousands of polyps - tiny sac-like marine invertebrates sea anemones and jellyfish, are important to marine ecosystems and prevent loss of life by acting as a natural barrier against storms, floods, and waves. These reefs support a wide range of species, many of which are underexplored and new species being discovered regularly. Crustose coralline algae (CCA) is one of the vital algal species that provides reef structure. Studying the abundance of CCA is important in helping marine biologists analyze coral reef health while understanding the impact of climate change on the marine lifeforms. This study …

Multimodal Techniques For Malware Classification, Jonathan Jiang

Master's Projects

The threat of malware has remained a serious concern for computer networks and systems, highlighting the need for accurate classification techniques. This research adopted the structured nature of PE files incorporated with a multi-modal machinelearning approach, to classify malware types. Features extracted from the PE headers were used to train an LSTM model. Features extracted from the PE sections were used to train a CNN model. Probabilities produced from these two models were then concatenated and fed into an SVM classifier. This multi-modal approach demonstrated high accuracy by experimenting with and verifying the approach on a large and labeled dataset. …

An Attributed And Diverse Encoder-Decoder Processing Technique For Anomaly Detection., Kenneth Antony John

Master's Projects

Attributed graphs are graphs that contain extra information about the attributes of nodes and edges. They can be used to model a plethora of real-world scenarios like social networks, bank transactions, and even academic citation data. Anomalies in such graphs can be irregularities or unusual patterns that are observed in the attributes or the structure of the graph. Anomaly detection in attributed networks is a crucial task, aiming to identify such anomalies. Existing methodologies use various deep learning techniques using graph neural networks, graph encoder-decoder architectures, and multi-layer perceptions. This study proposes a new approach to improve the existing methods …

Explainable Reinforcement Learning For Network Routing Optimization, Yu Xiu

Master's Projects

Software-defined Networking (SDN) provides a solution for configuring multiple network devices by offering a centralized controller architecture. Network routing is one of the most crucial problems in network configuration. In particular, the surging network traffic demands require efficient routing techniques to load balance communication links. In order to optimize the communication path’s utilization and reduce request blocking, this project utilizes Reinforcement Learning (RL) to decide the routes for given network requests. Furthermore, we adopt Explainable Reinforcement Learning (XRL) to explain the RL learning agent’s decision-making process to enhance the trustworthiness of our approach. In particular, we focus on Feature Importance …