arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2411.04434cs.LGcs.AI
收藏

智能体与世界模型的预训练缩放定律

Scaling Laws for Pre-training Agents and World Models

Microsoft Research(微软研究院)

Tim Pearce, Tabish Rashid, Dave Bignell, Raluca Georgescu, Sam Devlin, Katja Hofmann

AI总结:

本文研究了具身智能体预训练中的缩放定律,发现世界建模和模仿学习存在与语言建模类似的幂律,但定律系数受分词器、任务和架构显著影响,对模型与数据的最优规模设定具有重要指导意义。

摘要:

已证明具身智能体的性能可以通过增加模型参数、数据集规模和计算量来提升。当在离线数据集上使用生成式学习目标(预训练)来建模智能体的行为(模仿学习)或其环境(世界建模)时,这在从机器人到电子游戏的领域中得到了证明。本文更精确地刻画了规模在这些任务中的作用。超越“越大越好”的简单直觉,我们展示了在语言建模中发现的同类幂律也出现在世界建模和模仿学习中(例如损失与最优模型大小之间)。然而,这些定律的系数受分词器、任务和架构的严重影响——这对模型和数据的最优规模设定具有重要意义。

英文摘要:

The performance of embodied agents has been shown to improve by increasing model parameters, dataset size, and compute. This has been demonstrated in domains from robotics to video games, when generative learning objectives on offline datasets (pre-training) are used to model an agent's behavior (imitation learning) or their environment (world modeling). This paper characterizes the role of scale in these tasks more precisely. Going beyond the simple intuition that `bigger is better', we show that the same types of power laws found in language modeling also arise in world modeling and imitation learning (e.g. between loss and optimal model size). However, the coefficients of these laws are heavily influenced by the tokenizer, task \& architecture -- this has important implications on the optimal sizing of models and data.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑