Manu Mathew

About

Manu Mathew is a leading researcher in efficient deep learning hardware and algorithm design, with a primary focus on enabling high-performance Convolutional Neural Network (CNN) inference on resource-constrained embedded devices. His work addresses the critical challenge of deploying complex image classification models across automotive, industrial, medical, and robotics applications. Mathew’s major contributions include pioneering dynamic and predictive quantization techniques for CNN inference, which minimize memory and computational overhead without sacrificing accuracy. He also designed a groundbreaking VLSI architecture capable of achieving 1.2 TOPS (tera-operations per second) for convolution layers, setting a benchmark for on-device processing speed. Additionally, his development of an efficient frequency-domain CNN algorithm offers an alternative pathway to reduce latency in real-time systems. With his most-cited papers accumulating over a dozen citations, Mathew’s research has directly influenced the practical deployment of deep learning in edge computing. His work is notable for bridging the gap between theoretical neural network performance and real-world hardware constraints, making him a key figure in the advancement of efficient, scalable AI for embedded systems.

Research Focus

Key Achievements

Top Papers

1. 1
   CNN Inference: Dynamic and Predictive Quantization

   6 citations · 2018
2. 2
   CNN inference: VLSI architecture for convolution layer for 1.2 TOPS

   6 citations · 2017
3. 3
   Efficient frequency domain CNN algorithm

   2 citations · 2017

Key Collaborators

Contact & Links