Research Projects
Open-source tools and frameworks for Explainable AI, Video Understanding, and Cloud AI Systems
Featured Projects
Top Venues
STAA - Spatio-Temporal Attention Attribution
A real-time XAI method for interpreting video Transformer models. STAA extracts both spatial and temporal explanations simultaneously in a single forward pass, with less than 3% computational overhead. Achieves superior faithfulness (0.87) and monotonicity (0.91) scores on Kinetics-400 dataset with sub-150ms latency for real-time applications.
XAIport - Early XAI Adoption Framework
A microservice-based framework for integrating Explainable AI into ML development pipelines. XAIport shifts explainability from post-hoc analysis to an integral development practice. Compatible with Azure Cognitive Services, Google Cloud Vertex AI, and AWS Rekognition. Improves both model performance and explanation stability.
Joint Spatio-Temporal Adversarial Attack
A novel adversarial attack framework targeting video Transformer models through XAI-guided gradient perturbation. First method to simultaneously perturb spatial and temporal features. Achieves highest Attack Success Rate (ASR) on Kinetics-400, significantly exceeding V-BAD and sparse attacks. Reveals systematic security vulnerabilities in video transformers.
Cloud XAI - Trustworthy Cloud AI Explanations
An open API architecture for discovering trustworthy explanations of proprietary cloud AI services (Azure, GCP, AWS) without accessing internal model parameters. Provides feature contribution explanations with provenance data for full reproducibility. Enables data augmentation optimization through XAI-guided analysis.
More Projects
XAIpipeline - XAI Service Orchestration
An automated toolchain for orchestrating Explainable AI services across cloud and open-source models. Provides unified Open APIs, SDK, and web portals for configuring complex multi-step XAI workflows. Features CI/CD deployment support and comprehensive provenance tracking for reproducibility.
Interested in Collaboration?
I'm always open to discussing research collaborations, especially in Explainable AI and Video Understanding.