Kaichiu Wong

About

Kaichiu Wong is a researcher advancing the frontier of reinforcement learning (RL) by rethinking how machines learn from human input. Their primary research focus lies in reward learning and human-in-the-loop AI, specifically addressing the limitations of traditional preference-based RL. Wong’s most influential work, "Weak Human Preference Supervision for Deep Reinforcement Learning" (2021, 58 citations), challenges the conventional reliance on fixed, binary preferences between trajectory segments. They argue that human judgment is inherently dynamic and context-dependent, proposing a more flexible framework that accommodates weaker, less rigid supervision signals. This contribution is pivotal for scaling RL to complex, real-world tasks where defining a perfect reward function is impractical. In earlier work, "Human Preference Scaling with Demonstrations For Deep Reinforcement Learning" (2020), Wong explored integrating demonstrations to calibrate and scale human preferences, further bridging the gap between human intuition and algorithmic learning. Their research has significant implications for robotics, autonomous systems, and interactive AI, where safe and adaptable behavior is critical. By tackling the fundamental challenge of aligning AI with nuanced human values, Kaichiu Wong is helping shape a future where machines learn not just from data, but from the subtle, evolving preferences of the people they serve.

Research Focus

Key Achievements

Top Papers

1. 1
   Weak Human Preference Supervision for Deep Reinforcement Learning

   58 citations · 2021
2. 2
   Human Preference Scaling with Demonstrations For Deep Reinforcement Learning.

   2 citations · 2020

Key Collaborators

Contact & Links