🛠 Projects

Here are some of the projects I have worked on. For more up to date information on what I am working on, be sure to checkout my GitHub.

🛡 Adversarially Robust CNNs & LSTMs

CNNs, LSTMs, Adversarial Machine Learning

March 2023

bliutech/adversarial-networks

Exploring defenses against adversarial attacks on CNN and CNN-LSTM neural networks trained on CIFAR-10.

Details
  • Explored the impact of adversarial examples against common deep learning classification architectures (CNNs and RNNs) using CIFAR-10 dataset
  • Implemented Fast-Gradient Sign Method (FGSM) to generate adversarial perturbations against baseline models
  • Developed robust networks using data augmentation with adversarial examples to defend against attacks
  • Baseline models achieved 80.374% accuracy (CNN) and 79.001% (CNN-LSTM), but dropped to 9.375% and 12.679% respectively when attacked with adversarial examples (ε = 0.1)
  • Robust models trained on ε = 0.2 examples achieved 75.488% (CNN) and 74.056% (CNN-LSTM) validation accuracy under attack - an absolute increase of 64.142% and 61.375% respectively
  • Found that CNN architecture outperformed CNN-LSTM hybrid in both accuracy and training time for adversarial examples

💻 NLP-based Malware Detection of PDFs

Natural Language Processing, Transformers, Python

November - December 2022

bliutech/nlp-pdf-malware-detection

Static analysis technique for detecting malware in PDF files using natural language processing (NLP) techniques and transformer-based machine learning models, achieving 96.67% classification accuracy.

Details
  • Addressed the threat of malware hidden in PDFs through purposefully-embedded JavaScript
  • First exploration of using transformers for static PDF malware analysis, leveraging their attention mechanisms and parallel processing capabilities
  • Preprocessed PDFs as byte strings and generated word embeddings using one-hot encoding and variable n-grams
  • Fine-tuned transformer model to classify PDFs as malicious or benign with 96.67% accuracy on testing set
  • Demonstrated feasibility of transformer-based static analysis for PDF malware detection
  • Identified opportunities for further refinement and improved precision on varied datasets

📕 Wikisafe

React.js, Flask, Solidity (Ethereum)

October 2022

bliutech/wikisafe

HackMIT 2022. 🏆 2nd Place in Blockchain for Society (sponsored by Jump Crypto). Blockchain-based crowdsourcing platform that uses Ethereum to validate article contributions and prevent vandalism while enhancing collaboration with ML features.

Details
  • Leverages Ethereum blockchain to validate contributions to crowdsourced articles and prevent vandalism
  • Provides machine learning models to automate summarization, captioning, and figure generation
  • Improves overall contributor experience through intelligent automation
  • Simple account-based system for contributors to start sharing knowledge
  • Enables communities to collectively improve free and accessible education globally with enhanced security and quality

♻️ Recycling Elevated (Re)

React.js, Flask, PyTorch, Computer Vision

April 2022

bliutech/re

LA Hacks 2022. Interactive web application that uses computer vision to identify trash items and determine which recycling bin they belong to.

Details
  • Photo-based trash classification system that identifies the correct recycling bin for waste items
  • Built with React.js frontend and Flask backend powered by PyTorch computer vision models
  • User account system to track recycling activity and scanning history
  • Global leaderboard featuring users who scan the most trash items
  • Gamifies recycling to encourage environmentally responsible behavior

🫀 OrganSafe

React.js, Flask, Solidity (Ethereum)

February 2022

bliutech/organ-safe

TreeHacks 2022. 🏆 Grand Prize in Healthcare Winner. Blockchain-based organ donation platform that uses Ethereum to prevent black market trafficking and ensure secure, fair allocation of donated organs.

Details
  • Tackles the growing health and security problem of black market organ trafficking
  • Leverages Ethereum blockchain to verify organ recipients and prevent improper allocation
  • User registration system for donors to input health information and donation preferences
  • Automated matching algorithms to prioritize and allocate available donations fairly
  • Hospital tracking system to monitor organ donations and record when recipients receive transplants
  • Provides critical security for one of healthcare's most vital resources