Research
Work & Interests
Research assistant at CMATER Lab, Jadavpur University. Broadly interested in multimodal learning, efficient deep learning, and AI for embodied systems.
Publications
Wilson-Prime Channel Attention and Gini-Adaptive Dual-Backbone Fusion of Deep Models for Medical Image Classification
Sajjad Ahmed Shaaz et al. — CMATER Lab, Jadavpur University
Proposes a dual-backbone architecture (MobileNetV2 + DenseNet121) with novel Wilson Prime Channel Attention, Prime-Gini Adaptive Fusion, and linear-complexity global context modeling for medical image classification.
read more →
Multimodal Deepfake Detection via Audio-Visual Causal Divergence
Sajjad Ahmed Shaaz — CMATER Lab, Jadavpur University
Frames deepfake detection as identifying broken causal links between audio and visual modalities. Uses WhisperX forced phoneme alignment, CLIP ViT encoders, and Riemannian manifold representations to measure cross-modal synchrony divergence.
read more →
Research Areas
Multimodal Deepfake Detection
Investigating audio-visual causal divergence as a signal for deepfake detection. Current work explores cross-modal synchrony via WhisperX forced phoneme alignment, CLIP ViT encoders, and Riemannian manifold representations of causal link integrity between modalities.
Medical Image Classification
Dual-backbone deep learning architectures with attention mechanisms for robust medical image classification across pathology, histology, and radiology domains.
On-Device AI & Model Compression
Deploying deep learning models on constrained hardware via structured pruning, quantization, and knowledge distillation. Current focus on FPGA deployment with Vitis AI and TADNet for task-aware open-vocabulary detection.
Autonomous Systems & Robotics
Computer vision pipelines and reinforcement learning for autonomous drone and rover systems. Includes GPS-resilient navigation via IMU-based dead reckoning and RL-based manipulator control.
Interested in collaborating? Reach out.