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 2301.02679cs.LGcond-mat.dis-nn
收藏

Grokking模运算

Grokking modular arithmetic

Meta AI;Meta Platforms, Inc.(Meta平台公司);University of Maryland(马里兰大学)

Andrey Gromov

摘要:

我们提出了一个简单的神经网络,可以学习模块化的算术任务,并表现出突然跳跃的泛化称为“grokking”。具体地说,我们提出了(i)完全连接的两层网络,在没有任何正则化的情况下,在具有MSE损失函数的香草梯度下降下,在各种模块化算术任务上表现出grokking;(ii)grokking模块化算术对应于学习其结构由任务确定的特定特征图的证据;(iii)权重的解析表达式--因此也是特征图的解析表达式--解决了一大类模块化算术任务;以及(iv)这些特征图也通过香草梯度下降以及AdamW找到的证据,从而建立由网络学习的表示的完全可解释性。

英文摘要:

We present a simple neural network that can learn modular arithmetic tasks and exhibits a sudden jump in generalization known as ``grokking''. Concretely, we present (i) fully-connected two-layer networks that exhibit grokking on various modular arithmetic tasks under vanilla gradient descent with the MSE loss function in the absence of any regularization; (ii) evidence that grokking modular arithmetic corresponds to learning specific feature maps whose structure is determined by the task; (iii) analytic expressions for the weights -- and thus for the feature maps -- that solve a large class of modular arithmetic tasks; and (iv) evidence that these feature maps are also found by vanilla gradient descent as well as AdamW, thereby establishing complete interpretability of the representations learnt by the network.

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。

↑