Shankara Narayanan Venkateswara Raju

Optimizing on-device inference

Developed a pipeline using the Vision Transformers (ViT) to classify fetal ultrasound scans as good or bad quality and extract GradCAMs. Utilized these GradCAMs and images to cluster clinically relevant features. Built an assistive and interpretive ultrasound fetal scanning system.

• Worked on a Sony-funded project to build a smart and cost-efficient solar drying system. • Developed Edge-AI-based quantization-aware classification model for Sony Spresense board with 92% accuracy. • Performed real-time regression analysis (0.88 R2 score) to predict drying time. • Deployed solution achieving 76% real-time accuracy in classifying tomato dryness stages. • Integrated model with a 2-node Federated Learning pipeline for enhanced data privacy.

• Developed an ensemble model analyzing the System calls, Network traces, and Hardware Performance Counters for malicious behavior. Achieved an 89% accuracy on the RaDaR dataset. • Developed a quantitative model for malware classification based on efficient system call feature representations with <1% False Positive Rate. • Developed a two-stage clustering model using malicious HTTP network traffic flows from the RaDaR dataset. Extracted a generalized HTTP-based network signature for deployment on network edge devices, enabling efficient monitoring of network traffic. Learn more about the open malware research initiative from: https://www.openmalwareresearch.com

Best outgoing student in The School of Computing 24’