Changhyeon Kim

About

Changhyeon Kim is a leading researcher in energy-efficient deep learning hardware, with a focus on enabling real-time artificial intelligence for mobile autonomous systems. His work bridges the critical gap between the computational demands of deep reinforcement learning (DRL) and the stringent power constraints of edge devices like robots, drones, and autonomous vehicles. Kim’s most influential contribution is the development of a 2.1 TFLOPS/W mobile DRL accelerator, which introduced a novel transposable processing element array and experience compression techniques. This design, published in 2019 and garnering 59 citations, directly addresses the challenge of real-time operation in action control tasks, moving beyond traditional object recognition. His subsequent 2021 work further tackles the three major bottlenecks of DRL training acceleration, emphasizing the need for human-level adaptation in rapidly changing environments. By pioneering hardware that efficiently supports both inference and the computationally intensive training loops of DRL, Kim is laying the foundational architecture for truly autonomous, self-learning systems that can operate independently in the field.

Research Focus

Key Achievements

Top Papers

1. 1
   A 2.1TFLOPS/W Mobile Deep RL Accelerator with Transposable PE Array and Experience Compression

   59 citations · 2019
2. 2
   Energy-Efficient Deep Reinforcement Learning Accelerator Designs for Mobile Autonomous Systems

   4 citations · 2021

Key Collaborators

Contact & Links