Human - AI Interaction Loop Training as a New Approach for Interactive Learning

Explore a novel approach to interactive learning that combines Imitation Learning (IL) with Reinforcement Learning (RL) methods in this conference talk from ML Conference 2019. Dive into the challenges of using RL in complex decision-making tasks and discover how IL can leverage human-sourced assistance to overcome these obstacles. Learn about the proposed method that integrates IL with State-action-reward-state-action (SARSA) and Proximal Policy Optimization (PPO) algorithms to enhance agent learning in sequential decision-making policies. Examine the results of this innovative approach tested on various OpenAI-Gym environments, showcasing significant reductions in teacher effort and exploration costs. Gain insights into the potential of human-AI interaction loop training for improving machine learning outcomes in high-dimensional and continuous environments.