Papers
2
Total Citations
28
H-Index
2
About
Je Yang is a rising researcher at the intersection of deep reinforcement learning (DRL) and efficient hardware acceleration, with a focus on deploying intelligent agents in resource-constrained environments. His work addresses a critical bottleneck: the computational intensity of training and running DRL models, particularly for multi-agent systems used in robotics and autonomous driving. Yang’s major contribution is the development of **FIXAR**, a fixed-point deep reinforcement learning platform that integrates quantization-aware training with adaptive parallelism. This work, which has garnered 26 citations, demonstrates how to dramatically reduce memory and compute requirements without sacrificing model performance, making DRL viable for edge devices. Building on this, Yang introduced **LearningGroup**, a real-time sparse training method for FPGAs that uses learnable weight grouping to accelerate multi-agent reinforcement learning. This innovation tackles the unique challenge of coordinating multiple agents while maintaining low latency. Though early in his career, Yang’s focus on bridging algorithm design with hardware-aware optimization positions him as a key contributor to the next generation of efficient, deployable AI systems—essential for real-world applications like multi-robot control and autonomous vehicle fleets.
Research Focus
Key Achievements
Top Papers
- 1
- 2