BabyAI 1.1
David Yu-Tung Hui Mila, Université de Montréal Maxime Chevalier-Boisvert Mila, Université de Montréal Dzmitry Bahdanau Element AI Yoshua Bengio Mila, Université de Montréal CIFAR Senior Fellow
Abstract
The BabyAI platform is designed to measure the sample efficiency of training an agent to follow grounded-language instructions. BabyAI 1.0 presents baseline results of an agent trained by deep imitation or reinforcement learning. BabyAI 1.1 improves the agent’s architecture in three minor ways. This increases reinforcement learning sample efficiency by up to $3\times$ and improves imitation learning performance on the hardest level from $77\%$ to $90.4\%$ . We hope that these improvements increase the computational efficiency of BabyAI experiments and help users design better agents.
中文速览
BabyAI是一个用来测试智能体学习"听懂语言指令并执行"能力的实验平台，核心问题是如何用更少的训练样本让智能体学得更好。研究者在原有1.0版本基础上做了三处改动：去掉视觉编码器低层的最大池化、在FiLM层周围加入残差连接，以及将视觉输入从手工三元整数编码换成词袋式可学习嵌入（Bag-of-Words embedding）。这些改动看似细微，却将强化学习（Reinforcement Learning）的样本效率提升了最多3倍，并将最难关卡的模仿学习（Imitation Learning）成功率从77%提高到90.4%，还额外解锁了四个此前无法完成的关卡。这项工作的意义在于：它在几乎不增加计算成本的前提下大幅提升了基线性能，为后续在BabyAI平台上研究基于语言的智能体学习提供了更高效、更可靠的起点。
原文 arXiv:2007.12770;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2007.12770v1