Papers
2
Total Citations
92
H-Index
2
About
Minah Seo is a leading researcher in reinforcement learning (RL), specializing in credit assignment and sparse reward challenges that plague real-world robotic control. Her seminal 2019 work, "Rewards Prediction-Based Credit Assignment for Reinforcement Learning With Sparse Binary Rewards" (49 citations), introduced a novel framework that predicts future rewards to trace which actions truly drive delayed outcomes—a fundamental breakthrough for solving the credit assignment problem. This approach enables agents to learn effectively even when feedback is infrequent or binary. Building on this, Seo’s 2020 paper, "Sampling Rate Decay in Hindsight Experience Replay for Robot Control" (43 citations), tackled the rareness of successful experiences in high-dimensional state spaces. By dynamically adjusting the sampling rate of successful versus failed episodes, she enhanced the efficiency of hindsight experience replay (HER), allowing robots to master complex tasks with minimal reward signals. Her work bridges theoretical RL advances with practical deployment, directly impacting autonomous systems and robotics. With over 90 combined citations, Seo’s contributions are pivotal for students and engineers aiming to train agents in environments where rewards are scarce and delayed.
Research Focus
Key Achievements
Top Papers
- 1
- 2Sampling Rate Decay in Hindsight Experience Replay for Robot Control43 citations · 2020