Forging a Safer, Smarter Digital Future: From Deep Neural Insights to Trustworthy Social Interactions.

Research Highlight
We're hiring motivated Ph.D. students!
Our lab is at the forefront of designing revolutionary AI-centric solutions, championing both privacy and security, and reshaping our interaction dynamics with the surrounding world. Keen to be part of this pioneering group? Reach out with your details to sajedul.talukder AT or visit us at EGRA A0321.

SupremeLab @ SIU: Securing Tomorrow's Technology with Intelligent, Privacy-centric Solutions

At SupremeLab, we specialize in integrating sophisticated neural network models with privacy-centric technologies to create secure, private, and trustworthy digital environments. We focus on developing advanced cybersecurity solutions for critical infrastructure, particularly in the energy sector, including smart grid, DER and nuclear systems. Our expertise encompasses a broad spectrum, from federated learning, adversarial machine learning, and privacy-preserving machine learning to broader cybersecurity challenges. We excel in secure multi-party computation (MPC), on-device learning, and differential privacy, with a special focus on Federated Learning and Differentially-Private Machine Learning for sensitive sectors like healthcare. Addressing issues in distributed systems, especially in IoT and edge devices, we work on mitigating communication overheads, optimizing resource utilization, and safeguarding against cyber threats and privacy intrusions. Additionally, we harness machine learning for computer vision applications in technologies like UAVs, UGVs, cameras, and drones, all tailored to provide robust, privacy-enhanced security solutions for the smart ecosystem.

In addition, a significant part of our work at SupremeLab is reinforcing security and user privacy in online and geosocial networks. We are committed to designing, developing, and implementing a secure, trustworthy, and privacy-enhanced AI-driven online ecosystem, adhering to the principle of AI for social good. We leverage machine learning to develop next-generation social network tools, focusing on advanced abuse detection and defense mechanisms against fraud and abuse on platforms like Facebook, Twitter, Instagram, LinkedIn, and WeChat. Our goal is to ensure a safe, reliable, and ethically responsible online social interaction space, demonstrating the power of AI in promoting positive social change.


Media News | Oct 12, 2023

KMOV TV interviews Dr. Talukder

Dr. Talukder was contacted by KMOV (Channel 4), a prominent television station in St. Louis, Missouri, affiliated with CBS and MyNetworkTV to provide his expert opinion on cybersecurity and ransomware.

Publication News | Sep 15, 2023

Three papers accepted in ACM/IEEE ASONAM 2023

The 2023 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining, one of the top social computing conferences (acceptance rate ~13-18%).

Few Ongoing Projects

Tracking Emotional Sentiment Dynamics in Social Network Commentaries

Amidst the digital renaissance powered by social media's meteoric rise, a treasure trove of real-time public sentiment unfolds. Riding on this wave, we've created a cool new tool that dives into these chats and figures out how people feel. Harnessing the might of the avant-garde BERT model, we sculpt emotional symphonies from the raw, unfiltered voices of netizens, painting a panoramic view across multifaceted sectors.

Zero-Knowledge Proof & Blockchain Based Framework for Combating Fake Profiles

Imagine a digital bouncer for social networks, ensuring only real folks enter. Our brainchild marries zero-knowledge proof (ZKP) with private ledgers (blockchains) to check you're you—without peeking into your life! It's like having a government-stamped ID in the virtual world and a secret handshake for entry. By merging ZKP and private blockchains, we ensure user authenticity and confidentiality, enhancing online safety.

Combating Identity Misrepresentation through Text-Driven Gender Recognition

Navigating the realm of social media, ensuring genuine gender representation is key for a safe, inclusive online world. While many languages have been explored, Bangla's digital landscape remained untouched. We apply stylometric, TF-IDF, and word embedding to identify the author's gender from Bangla texts. Traditional models outperform deep learning models for all except the stylometric features with Stochastic Gradient Descent outperforming state-of-the-art.

Federated Learning Based Contraband Detection within Airport Baggage X-Rays

For tighter airport security, spotting contraband in X-rays is vital. Amid high stress, security teams often miss hidden items. Enter machine learning: while great at detecting forbidden items, it can risk data privacy. Our solution? A privacy-friendly Federated Learning approach using top-tier algorithms. With accuracy rates hitting a high of 90%, we demonstrate that our method isn't just safe but also super efficient!

Predicting Child Lead Toxicity with Machine Learning and NLP at ZIP Code Level

Childhood lead exposure harms health, with links to factors like income and older homes. To pinpoint exposure hotspots, we used data from Massachusetts and New York, forecasting at the zip code level. Adding sentiment analysis, we assessed how well lead programs work. Our techniques, especially LightGBM, proved highly accurate to predict exposure risks by zip code. We've surpassed past scores, showing a detailed lead exposure map!

Differentially Private Federated Learning for Diabetic Retinopathy Prediction

Deep learning aids diabetic retinopathy diagnosis using large retinal image sets. However, data privacy laws restrict sharing. We suggest a secure federated learning system for cross-organization image analysis. Testing on a 35,120-image dataset, ResNet50 topped with 83.05% accuracy without noise and 79.35% with noise utilizing differential privacy. Our method also cut communication overhead by 49% against standard federated learning.