This article presents a new AI learning framework, Communication-Efficient Interactive Learning (CEIL), which promotes progressively efficient communication. Traditional frameworks like imitation learning and reinforcement learning only support low-level, inefficient communication. In contrast, CEIL equips the learning agent with a dynamic language and an intrinsic motivation to learn with minimal communication effort. This leads to a human-like pattern where the learner and the teacher communicate progressively efficiently by exchanging increasingly abstract intentions. The performance of CEIL is demonstrated on a 2D MineCraft domain, where agents trained with CEIL quickly master new tasks, outperforming other learning methods.
Publication date: 23 Oct 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2310.13004
