Vimal William
Ph.D. Student, Electrical & Computer Engineering
University of Arizona
🏆 Herbold Fellow
About
I am a first-year Ph.D. student at the University of Arizona, co-advised by Prof. Jyotikrishna Dass and Prof. Ravi Tandon.
My research lies at the intersection of machine learning, computer architecture, and distributed systems, with a focus on designing domain-specific systems for AI and optimizing existing systems to improve AI efficiency.
Currently, I'm working on optimizing transformer-based AI architectures to efficiently handle longer context lengths, with an emphasis on reducing memory utilization and improving scalability for large-scale applications.
Research Interests
- AI Systems Architecture: Domain-specific hardware and systems for machine learning workloads
- Transformer Optimization: Efficient long-context processing and memory optimization
- Compiler Engineering: Graph compilers and optimization frameworks (MLIR, TVM)
- Distributed Systems: Scalable ML infrastructure and distributed training
Education
Ph.D. in Electrical Engineering
University of Arizona, Tucson, AZ
Advisors: Prof. Jyotikrishna Dass, Prof. Ravi Tandon
Herbold Fellow
B.E. in Electronics and Communication Engineering
Anna University, Chennai, India
Experience
System Software & Compiler Engineer
SandLogic Technologies, Bangalore, India
Worked on compiler optimization and system software for AI accelerators
Publications
Study on the Behaviour of Mel Frequency Cepstral Coefficient Algorithm for Different Windows
ICITIIT 2022
Comprehensive study on the effects of different windowing functions on the MFCC algorithm for speech signal processing and audio feature extraction.
Selected Projects
Tiny Graph Compiler
Minimal implementation of a graph compiler based on the MLIR stack, following approaches proposed by the TVM white-paper. Focuses on backend targeting NVPTX for GPU code generation.
MLIR CUDA NVPTX Compiler Design
VStream
GStreamer-based streaming pipeline for on-device object detection. Supports CPU/GPU INT8 inference with optimized performance for edge deployment scenarios.
GStreamer Object Detection Edge AI INT8 Quantization