Research Interests
- Privacy Policy
- Differential Privacy
- Large Language Models
- Conspiracy Theories
- Honeypots
- Addresses a critical privacy challenge by using Differential Privacy to create private, synthetic health data that protects individual identities.
- Successfully applied this method to a real-world health study (LEMURS) of college students, protecting sensitive data collected from Oura rings and personal health surveys.
- Demonstrates that a practical balance is achievable, identifying a "sweet spot" where the generated data remains highly useful for researchers while significantly reducing privacy risks for participants.
- Developed a scientifically structured 'family tree' of conspiracy theories, categorizing and illustrating the connections among various conspiracies to enhance community understanding.
- Created the dataset by scraping articles from fact-checking websites and efficiently labeling them using Keyphrase Extraction, simplifying the process of identifying the main themes in each article.
- Developed a binary classifier using various machine learning methods, and our RoBERTa model achieved the highest performance with an F1 score of 87%, effectively distinguishing between conspiracy-related and non-conspiracy articles.
- Utilized the HDBSCAN + UMAP algorithm to facilitate effective data clustering and exploration, generating labels to be added to the main family tree.
Hacker Detector with Honey Documents
- Developed Google documents filled with simulated hacking methods for distribution acrosspaste sites.
- Utilized Cutlly API, Google App Script, and a self-controlled domain to examine visitor metrics such as visit count, edits made, geolocation, browser type, operating system, and device used.
- Differentiated between bots and non-bots accesses.
Performed support vector machine (SVM), multilayer perceptron (MLP), convolutional neural network (CNN), decision trees, k-fold cross validation, linear and logistic regression on metadata of over 1 million songs to classify the genre, based on 7 different features.
- Simulated a secure, distributed cryptocurrency system, ensuring transparency and trust by enabling each participant to control and validate the ledger, effectively preventing fraudulent activities such as double spending.
- Leveraging a robust peer-to-peer network framework, our simulation demonstrates the essential processes of a cryptocurrency operation including transaction signing, block mining, broadcasting, and validation, culminating in a dynamic and decentralized digital currency ecosystem.