Research Projects
Selected research highlights spanning robust artificial intelligence, vision-language models, federated learning, multimodal systems, and trustworthy machine intelligence.
Vision-Language Models Federated Learning Robust, Efficient and Secure AI Trustworthy AI, Agentic AI and Cybersecurity Real-World Applications
Do LVLMs Know When They Are Wrong? Trajectory-Based Correctness Estimation from a Single Forward Pass
Trajectory-Based Correctness Estimation for Large Vision-Language Models
We introduce a trajectory-based framework for estimating whether a Large Vision-Language Model's response is correct using internal reasoning dynamics from a single forward pass, enabling efficient correctness estimation without repeated model queries.
When Data is Scarce, Learn to Adapt: Robust Federated Learning via Adversarial Meta-Optimization
FAML
We propose FAML, the first and robust FAT framework that leverages meta-learning to enhance robustness in federated learning and address the challenges posed by data scarcity in heterogeneous clients.
Research Architecture Overview
Our novel defense mechanism protects Vision-Language Models against dual exploits through robust encoding and adversarial training techniques.
Sim-CLIP: Unsupervised Siamese Adversarial Fine-Tuning for Robust Vision-Language Models
Research Architecture Overview
Unsupervised approach to enhance Vision-Language Models through Siamese adversarial fine-tuning for improved robustness and semantic richness.
Towards Trustworthy Autonomous Vehicles with Vision-Language Models Under Adversarial Attacks
Research Architecture Overview
Examining the robustness of Vision-Language Models in autonomous vehicle applications under targeted and untargeted adversarial attacks.
Research Architecture Overview
Blockchain-enhanced framework ensuring privacy and security in distributed machine learning across edge computing environments.
Research Architecture Overview
Comprehensive defense mechanism for Vision-Language Models focusing on robust encoding techniques against various attack vectors.
TriplePlay: Enhancing Federated Learning with CLIP for Non-IID Data and Resource Efficiency
Research Architecture Overview
TriplePlay, a framework that tailors CLIP foundation model as an adapter to strengthen FL model’s performance and adaptability across heterogeneous data distributions among the clients.
Quantifying Robustness and Sustainability Trade-off in Federated Adversarial Learning for Cyber-Physical Systems
Research Architecture Overview
Federated Adversarial Learning simulation with nine heterogeneous devices with the different configurations.
Research Architecture Overview
Securing Privacy in Cloud-Based Whiteboard Services Against Health Attribute Inference Attacks.
Research Architecture Overview
Research Architecture Overview
Overview of experimental process: learning from hand drawing dataset, real-time drawings from DP-WhiteBoard tool, and recognizing shape and inference via transfer learning models.
Towards Resilient Critical Infrastructure Operations against Natural Calamity
Research Architecture Overview
Visualization of fire and hurricane occurrence event on map (H: Hurricane F: Wildfire).