Yoshihiro Mitsuka

About

Yoshihiro Mitsuka is at the forefront of making intelligent systems both more transparent and more resilient. His research primarily bridges deep reinforcement learning (DRL) and robotics, with a critical focus on explainability and hardware fault tolerance. Mitsuka’s most influential work, "Explainability of deep reinforcement learning algorithms in robotic domains by using Layer-wise Relevance Propagation" (2024, 9 citations), directly addresses the "black box" problem of neural networks in robotics. By applying Layer-wise Relevance Propagation, he provides a method to visualize and understand the decision-making processes of DRL agents, a crucial step for deploying safe and trustworthy autonomous systems. Complementing this, his work on "Enhancing Hardware Fault Tolerance in Machines with Reinforcement Learning Policy Gradient Algorithms" (2024, 2 citations) pioneers a novel approach: using reinforcement learning itself to dynamically adapt to and overcome physical hardware failures, moving beyond traditional, static redundancy methods. Through these contributions, Mitsuka is not only advancing the theoretical foundations of DRL but also paving the way for more robust and interpretable robots capable of operating reliably in the real world.

Research Focus

Key Achievements

Top Papers

1. 1
   Explainability of deep reinforcement learning algorithms in robotic domains by using Layer-wise Relevance Propagation

   9 citations · 2024
2. 2
   Enhancing Hardware Fault Tolerance in Machines with Reinforcement Learning Policy Gradient Algorithms

   2 citations · 2024

Key Collaborators

Contact & Links