Yukihide Kohira

About

Yukihide Kohira is a researcher whose work lies at the intersection of hardware design and machine learning acceleration, with a particular focus on energy-efficient computing. His key research areas include approximate computing, convolutional neural network (CNN) accelerators, and low-power VLSI design. Kohira’s most notable contribution is the development of a random-forest-based approximation layer unit (RFA-LU) for binary and ternary CNNs, as detailed in his 2021 paper. This work addresses the growing demand for faster, smaller, and low-power accelerators essential for real-time robotics control and reduced network load. By leveraging random forest techniques for approximation, his approach achieves significant area efficiency without sacrificing inference accuracy. Though his most-cited paper currently holds 4 citations, the innovative nature of his work positions it as a foundational contribution to the field of approximate neural network hardware. Kohira’s research is particularly impactful for students and engineers seeking to bridge the gap between algorithmic efficiency and practical hardware constraints, offering a promising pathway toward deploying advanced AI in resource-constrained environments.

Research Focus

Key Achievements

Top Papers

1. 1
   Area-efficient Binary and Ternary CNN Accelerator using Random-forest-based Approximation

   4 citations · 2021

Key Collaborators

Contact & Links