Digital Commons Network^™

Equity In Public Budgeting: Community Engagement In Morgan Hill, Ariana Perez

Master's Projects

Through this research, I will examine what citizen participation strategies have been implemented and their role and effectiveness in addressing wealth inequities. My research will explore these topics and examine what programs and policies the City of Morgan Hill can implement to increase residents' sense of belonging and ultimately push for greater social equity. My primary research question is, what are the programs and policies that the City of Morgan Hill can implement to engage immigrant communities, specifically Spanish-speaking residents, and low-income residents in shaping decisions around public investments and funding? This project will serve as a needs assessment to …

Homomorphically Encrypted Faceted Values, Tanmay Singal

Master's Projects

Faceted values prevent the implicit flow of sensitive information by controlling the visibility of program data. They achieve this by maintaining two facets for each variable: a public facet, which is observable, and a private facet, which remains hidden. Although this method secures the flow of sensitive data, it can be leaked if the server storing the faceted values is compromised. While faceted values may be encrypted on the server, doing so would necessitate that the private facets be briefly decrypted during execution to allow arithmetic operations to be performed on them, creating an attack vector for information to be …

Bot Detection In Social Media Using Graphsage And Bert, Abhishek Deshmukh

Master's Projects

This project details a novel bot detection system developed to battle the ever- changing challenge of disinformation, misinformation, and other bot-generated content.

The methodology employed in this project combines the text-based analytical strength of BERT (Bidirectional Encoder Representations from Transformers) with the strength of GraphSage (Graph Sample and Aggregation) for analyzing network structures. The project concatenates BERT and GraphSage vectors to create an 896-size feature embedding with a rich blend of network and text features. This project employs a Support Vector Machine to process the concatenated embeddings, as SVM works well with high-dimensional data. This project was evaluated on two …

Ai Powered Legal Decision Support System, Alisha Rath

Master's Projects

The large volume of legal cases presented by judicial professionals has made it
challenging to study and predict results. With advances in research methods and
technology, predicting law cases in a more accurate manner has become an important
trend. Prediction tools based on AI may help manage a large number of legislative
texts and documents that cannot possibly be fully read, reduce the number of cases
to be seen, and give accurate outcomes of how cases may turn out. Now, when
we look into the current AI legal prediction tools in this domain, they mostly lack
efficiency and interpretability, the …

Multimodal Feature Fusion And Machine Learning For Adhd Detection Using Neuroimaging Data, Isabel Pham

Master's Projects

Attention Deficit Hyperactivity Disorder (ADHD) is a common neurodevelopment disorder that can significantly affect a person’s attention, impulse control, and executive function. Currently, the traditional diagnosis method often relies on clinical assessments and observations. However, these methods can be subjective and lead to inconsistencies in diagnosis between individuals. To address this challenge, neuroimaging and machine learning (ML) are promising tools for providing a more objective diagnosis of ADHD. The goal of this project is to apply a multimodal approach in which structural and functional features of specific regions of the brain are used to develop a more accurate and objective …

Machine Learning Based Network Traffic Classification With Cosine-Similarity Based Out-Of-Distribution Detection, Prabhat Edupuganti

Master's Projects

The changes occurring in the amount of encrypted network traffic is growing at an alarming rate. This development has created intricate problems in traffic classification which is vital for effective cybersecurity. Moreover, most frameworks seem to ignore OOD detection, model calibration and novel pattern detection as cornerstone problem areas. The due analysis is presented as a machine learning approach aimed at resolving encrypted traffic classification issues and focuses on novel OOD detection and calibration issues. Primary contributions comprise detection of out-of-distribution states using softmax scaled cosine similarity, advanced variance-based feature elimination, and lowering ECE using stringent NNs. This work demonstrates …

Mitigating Cold Start Problem Through Metadata Integration And User Preference Analysis, Prabaljit Walia

Master's Projects

Recommendation systems power the most popular platforms in the world: from content catalogs on Netflix to custom feeds on TikTok – the importance of recommendation systems is significant. Collaborative filtering, the most popular recommendation technique, is essentially based on the idea of leveraging collective user intelligence i.e., creating recommendations by finding similar users. But this technique suffers when there is not enough data in the profiles of users, formally termed as the cold start problem. This research focuses on this problem by introducing an approach that integrates metadata-driven similarity measures with profile expansion techniques. Our approach combines traditional collaborative filtering …

Real-Time Adaptive Framework For Topic Modeling In Social Engineering Attacks, Manav Bhasin

Master's Projects

Detecting social engineering attempts is crucial for security, as these threats are becoming more frequent and increasingly exploit human vulnerabilities. This research focuses on topic modeling using conversational data from Kevin Mitnick’s ”The Art of Deception” with dialogues that illustrate various social engineering strategies. The dataset comprises manually extracted and synthetically augmented conversations to ensure natural dialogue flow. Two methodologies are presented for utterance-level and global topic extraction: prompt engineering leveraging OpenAI’s GPT-4o-mini, characterized by few-shot learning and chain-of-thought prompting, and Quantized Low Rank Adaptation (QLoRA) utilizing Mistral’s 7B instruct model for efficient fine-tuning. Through experimentation and evaluation, this study …

Ai-Enabled Anticipatory Handover Predictions In 5g Networks, Ojas Ankush Naik

Master's Projects

As users move across network cells in 5G, it is critical to maintain seamless connectivity through efficient and fast handovers. However, as 5G networks have a very dense deployment of cells and higher carrier frequencies, handovers are often more frequent and challenging, leading to failures or the ping-pong effect. In this research, we are going to use Artificial Intelligence (AI) techniques to enable predicting handover(HO) events proactively as opposed to reactively, aiming to reduce HO failures and unnecessary handovers. We develop Long-Short-Term Memory (LSTM) and Bidirectional LSTM (Bi-LSTM) models to forecast future signal measurements, predict handover trigger points, and compare …

Simple Vs. Complex Human Activity Classification Via Hybrid Machine Learning Models, Anusha Kukreja

Master's Projects

In-depth understanding of the complexity of daily human activities is crucial for building responsive health monitoring and assistive technologies. However, limited research has focused on distinguishing activities based on their involvement level, as most existing work classifies only the type of activity performed. In this thesis, we address this gap by proposing a method to classify human activities as either simple or complex using sensor data from the Opportunity dataset. We define complex activities as those involving object interactions or multiple coordinated movements (e.g., drinking from a cup, cleaning a table), and simple activities as static or low-effort postures (e.g., …

Visionmate: Ai-Powered Image Captioning Web Application, Sai Anoushka Kokku

Master's Projects

VisionMate is a web application that generates captions for camera-captured images. It is designed
to assist users with visual impairments by converting visual input into spoken and written text. The
application uses the GIT-base model from Hugging Face, which processes the image and returns a
descriptive caption. Users can take a picture using the device camera—either via webcam on
desktop or the native camera interface on mobile. The app provides audio output using the
SpeechSynthesis API and uses full-screen tap interaction to simplify accessibility.

The frontend is implemented in React.js, and the backend is built with FastAPI. The backend calls …

Adaptive Cobot Interaction Via Smartwatch Data Fusion For Car Assembly Automation, Riddhik Tilawat

Master's Projects

In modern car manufacturing, collaborative robots (cobots) work with human operators during shared workcell interactions to maximize production speed and flexibility. Collaboration between humans and robots is safe and effective only when operator intent recognition via a single wrist-worn inertial measurement unit (IMU) is accurate and low-latency. This thesis develops an IMU-only intent recognition pipeline, and is evaluated on three datasets: the public OPPORTUNITY dataset, the Sony Smartwatch Gesture dataset and a custom Samsung Galaxy watch 6 dataset. The proposed framework leverages five step sequence-to-label problems which are stepwise posed as data streams transforming raw IMU data into trainable tensors. …

Augmenting Missing Sensor Data For Robust Human Activity Recognition, Suryakangeyan Kandasamy Gowdaman

Master's Projects

Applications of ubiquitous computing, including health monitoring, sports analytics, and ambient-assisted living, rely on Human Activity Recognition (HAR) using wearable sensors. However, model robustness is challenged by missing sensor values, class imbalance, inter-subject variability, and temporal noise. This work proposes a complete HAR pipeline that addresses these challenges through sampling, time-series augmentation, dynamic feature handling, and GAN-PCA-based imputation. Built on the DeepSense architecture, the model integrates convolutional feature extraction with bi-GRUs for temporal modeling. The system is evaluated using 5-fold cross-validation, subject-aware holdout, and LOSEO strategies on the Opportunity dataset. Results demonstrate consistent accuracy across folds and strong generalization to …

Faux Capabilities: A Novel Approach For Code Analysis, Tanay Godse

Master's Projects

When using third-party packages or libraries, it is crucial to understand their behavior. Typically, this requires developers to either conduct code reviews or set up sandbox environments for testing or write unit tests with mocked values for every function used in their code. However, these approaches are often inefficient and time-consuming. A more effective solution would provide developers with a broad understanding of the functionality required by the code they plan to import. This can be done using object capabilities, where a particular functionality is the capability that an object must possess, in order to be able to perform the …

Traceai: Intelligent Distributed Tracing Using Large Language Models, Mihir Dhirajlal Satra

Master's Projects

Distributed systems are difficult to trace using traditional methods due to the scale of data volume and complexity, and they usually require a lot of manual analysis. TraceAI tries to solve these problems by integrating Large Language Models with the tracing tools to automatically enhance the trace data evaluation. The project aims to provide an AI-driven solution for monitoring and understanding the flow of requests across services, anomaly detection, root cause analysis and performance optimization. It can thus automate finding out systems problems using LLMs thereby carrying out large scale trace data analysis. Anticipated results from the effort will be …

Unconditional-To-Conditional Transfer And Optimization For Web-Based Skybox Gan, Crystal Kwong

Master's Projects

Generative adversarial networks (GANs) are known for their ability to generate high quality
images mimicking real life or even particular art styles. Yet for all their capability, casually
training a GAN on an average machine can be infeasible as GANs require an enormous amount
of time and data to train. Even with a trained GAN, model inference demands heavy
computations, making GANs difficult to deploy on applications. To address these limitations,
techniques such as transfer learning and quantization have been leveraged to speed up training of
GANs and lighten computational cost of GAN inference. This project aims to use such …

Effects Of Data Augmentation On Sponge Identification Using Computer Vision Models, George Ku

Master's Projects

Coral reefs can be primarily found in tropical and sub-tropical regions of our oceans, providing a thriving habitat for millions of species. Marine sponges, which can be frequently found in coral reefs, play a critical role that contributes to the maintenance of these ecosystems, including the recycling of nutrients through water filtration. However, rising ocean temperatures and acidification due to climate change have resulted in the bleaching and death of coral reefs worldwide. In order to preserve these reefs and the sponges that depend on them, scientists have been performing studies on their biodiversity. This includes collecting numerous images of …

Proxy Cap - La Lua Protector, Swift Sheng

Master's Projects

Inspired by the object capability model and sandbox, this project, Proxy Cap, introduces a new Lua access control model that improves the language’s security without sacrificing usability. Object capability is an unconventional but powerful security model. The security model closely observes the principle of least authority. Ambient authority, the omnipresent global environment, does not exist in the object capability computation world, and no resource is accessible unless explicitly assigned. Only connectivity begets connectivity. Lua is an extensible and high-performing scripting language based on ANSI C. The language is popular in many fields but faces security challenges. Lua has a non-traditional …

Combining Esm Models With Experimentally Derived Structural Stability To Identify Functional Missense Mutations, Rucha Deo

Master's Projects

Missense mutations can impact protein function and structure, yet their effects on protein function are difficult to predict. In this study, I compared two deep learning models, ESM1v and ESM1b, by evaluating their mutation predictions against experimental structural stability data. ESM1v showed a stronger correlation with experimental structural stability scores compared to ESM1b. A sigmoid curve was fitted to explore this relationship further. Over 100,000 mutations were identified where experimental stability differed significantly from model predictions. Many mutations that remained structurally stable experimentally but were predicted as harmful by the ESM models were frequently found at known functional sites. Structural …

Ribomoe: An Application Of Mixture Of Experts On Artificial Riboswitch Classification, Hainian Audrey Long

Master's Projects

Urban water sewage is a potential health concern due to its possibility to spread contagious RNA viruses such as Coxsackievirus B3. However, detection of viral particles remains challenging because of low viral concentrations in wastewater and high mutation rates of the RNA virus. To address this, this study proposes a novel viral detection method using synthetic riboswitches that bind to the target virus and trigger a reporter gene, amplifying the detection signals. To support the design of effective riboswitches, we present a machine learning model for classifying riboswitch performance, integrating RNA sequence data with secondary structural features. This model used …

Phishing Detection Using Continual Learning And Large Language Models, Gopi Prajeev Battula

Master's Projects

Adaptive phishing detection remains crucial as the nature of cyber-attacks changes over time, which renders static models obsolete. This project extends phishing detection through the implementation of continual learning approaches, namely Elastic Weight Consolidation (EWC) and Learning Without Forgetting (LWF) with RoBERTa, a Large Language Model (LLM) and compares the results of these approaches against GPT-4o-mini, another LLM. Our approach begins with fine-tuning RoBERTa on multiple phishing datasets to establish an effective baseline. EWC is then implemented to preserve vital model parameters based on their importance measured by the Fisher Information Matrix, while LWF uses knowledge distillation to retain prior …

Disease Diagnosis Using Rag Llm With Smart Prompt Engineering, Qadeerullah Syed

Master's Projects

Although recent trends indicate that LLMs outperform traditional methods in solving complex problems with enhanced reasoning, there has been barely any progress in replicating the quality of diagnoses like those of actual human doctors. The identification of an accurate diagnosis with thorough reasoning is still a significant challenge, even with advanced AI models. The process of performing accurate diagnosis remains challenging due to a lack of transparency in state-of-the-art models existing today, a lack of explanation in the diagnosis process, an emphasis on results rather than reasoning, and a lack of foundational knowledge in models, along with limited exploration of …

Hierarchical Bloom Filter Tree (Hbft): Scalable Geospatial Metadata Indexing For Bigdata Systems, Mrudula Patteparapu

Master's Projects

Big Data infrastructure growth has produced overwhelming metadata volumes which create extensive problems for spatial indexing and both system scalability and database queries. Traditional solutions consisting of R-trees and conventional Bloom filters manage to provide either range query support or approximate membership testing, yet they face performance issues when used at large-scale metadata management. This research proposes Hierarchical Bloom Filter Tree (HBFT) as an improved framework that integrates hierarchical spatial partitioning with partitioned, scalable, cuckoo, and striped Bloom filter variants based on existing studies in hierarchical and probabilistic indexing. The complete evaluation process shows that HBFT outperforms PostGIS (an industry-standard …

Identifying Red Sponges On Arms Plates By Preprocessing Images Using Histogram Equalization, Barry Ng

Master's Projects

Sponges play a vital role in marine ecosystems, being the only organisms capable of converting dissolved organic matter (DOM) into particulate organic matter (POM). They provide nutrients for coral reefs to thrive in oligotrophic waters. Autonomous reef monitoring structures (ARMS) are used to measure the biodiversity of coral reefs by simulating the complex cavities inside reef structures. Organisms settle on them and scientists can retrieve them after a period of time for analysis. Images are taken of ARMS plates after they are retrieved. Human analysis is unsuitable for the analysis of ARMS plates due to the huge number of images. …

An Evidence-Based Approach To Predicting Pancreatic Ductal Adenocarcinoma, Surya Teja Nalluri

Master's Projects

Pancreatic ductal adenocarcinoma (PDAC) is a complex disease with hidden clinical indicators, so a reliable diagnosis of PDAC requires high precision and sophisticated analysis. Traditional probabilistic methods often rely on making unwarranted assumptions or undesirable approximations about probabilistic estimates, limiting their ability to provide the precision needed for correct diagnosis and treatment planning. In contrast, Dempster–Shafer Theory offers a formal framework for integrating uncertain and potentially conflicting evidence. This makes it well-suited for analyzing incomplete and ambiguous data typically associated with PDAC. By employing an evidential reasoning (ER) model based on Dempster-Shafer Theory, this approach systematically combines and evaluates imperfect …

Transformers In Time-Series Forecasting: Enhancing Robustness Via Dynamic Attention Mechanisms, Kush Patel

Master's Projects

Transformer architectures have emerged as powerful tools for time series forecasting, excelling at capturing complex temporal dependencies across multivariate inputs. However, these models are highly susceptible to adversarial attacks such as the Fast Gradient Sign Method (FGSM) and Basic Iterative Method (BIM), which can significantly degrade predictive performance through small, targeted perturbations. This work integrates dynamic attention mechanisms, adaptive masking modules that introduce controlled variability into attention pathways, into a transformer forecasting model to enhance robustness against such attacks. Using two distinct datasets, we compare the performance of a standard transformer and a dynamic attention-enhanced transformer under both clean and …

On The Adversarial Robustness Of Quantized Neural Networks Against Common Adversaries In Time-Series Forecasting, Maanak Arora

Master's Projects

Real-world edge applications now use modern machine learning models which require both resource efficiency and robustness against adversarial threats. Deep neural networks which include time series forecasting models still face risks from adversarial perturbations while quantization techniques used for memory and compute efficiency create unpredictable robustness challenges. This project investigates the adversarial resistance of Long Short-Term Memory (LSTM) models after applying post-training quantization at three different precision levels: 16-bit floating point (FP16), 8-bit integer (INT8) and custom 4-bit quantization. The Jena Climate dataset serves as our main benchmark for training a fullprecision LSTM model followed by multiple quantization strategies which …

Transformer Integration, Fine-Tuning And Zero-Shot Learning For State Of Health Estimation In Li-Ion Batteries Using Large Language Models, Chinmay Nilesh Mahagaonkar

Master's Projects

In this thesis, we present a comparative analysis of the use of transformerbased and Large Language Model (LLM) models for State of Health (SoH) and Remaining Useful Life (RUL) prediction of lithium-ion batteries. With electric cars and renewable energy systems based on batteries at the forefront, the need to predict degradation accurately in order to enhance the performance and reduce maintenance costs has become imperative. Most traditional prediction methods lag the complex and non-linear characteristics of degradation in batteries, and hence the usage of sophisticated methods becomes a necessity. The research employs the CALCE dataset, which includes long-horizon cycling data …

Medilightrag: A System For Medical Query Response Using Fine-Tuned Llms And Graph Based Retrieval, Rajiv Karthik Reddy Kodimala

Master's Projects

The exponential increase in medical data has created a greater demand for precise and efficient information retrieval systems. Existing Large Language Models (LLMs) face domain-specific difficulties such as sophisticated medical jargon, situational comprehension, and the continual advancement of healthcare knowledge. To tackle these challenges, we present MediLightRAG, an innovative two-stage system which integrates parameter-efficient fine-tuning of Large Language models with LightRAG’s graph-based retrieval. The first stage focuses on enabling accurate resource-efficient model adaptation for the medical domain through QLoRA fine-tuning. In the second stage, LightRAG’s two-tiered retrieval architecture that combines graph-based indexing with dynamic knowledge retrieval is employed to enhance …

Edurag: Improving Ai Teaching Assistants With Retrieval-Augmented Generation, Geethika Vadlamudi

Master's Projects

This paper is based on the emerging need for AI-driven teaching assistants to deliver personalized, effective, and responsive educational assistance. Earlier proposals for educational technology such as rule-based systems and adaptive learning platforms have been limited by the lack of flexibility, domain accuracy, and slow, non-interactive support. With the launch of large language models (LLMs) like GPT-3, GPT-4, they have demonstrated that they can generate the kind of human-responsive text. When it comes to more substantive education matters, though, such models suffer from domain accuracy, in addition to whether accurate information can be imparted. And that is where the aspect …