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 2606.01302cs.LG
收藏

单纯复形序列建模中的结构与规模

Structure and Scale in Simplicial Sequence Modelling

Department of Computer Science, University of Oxford(牛津大学计算机科学系)

Matthew Farrugia-Roberts

AI总结:

本文通过训练小型Transformer预测隐马尔可夫模型输出,发现性能缩放模式与内部表征之间存在相关性,为行为缩放定律与涌现机制的联系提供了初步证据。

摘要:

现代大规模深度学习展现出两个引人注目的经验现象:行为缩放定律(性能随规模增大而可预测地提升)和涌现机制(深度神经网络中结构化的内部表征和回路)。我们假设这两个现象是相互关联的:行为上的可预测变化是内部计算结构可预测变化的结果。在本文中,我们报告了这种联系的初步证据。我们发现,在训练用于预测隐马尔可夫模型输出的小型Transformer中,性能缩放模式与表征之间存在相关性,已知其残差激活在概率单纯形上线性编码了关于潜在状态的信念分布。

英文摘要:

Modern large-scale deep learning exhibits two striking empirical phenomena: behavioural scaling laws (predictable performance gains with increasing scale) and emergent mechanisms (structured internal representations and circuits in deep neural networks). We hypothesise that these two phenomena are connected: that predictable changes in behaviour are the result of predictable changes in internal computational structure. In this paper, we report preliminary evidence of such a connection. We find a correlation between scaling patterns in performance and representations in small transformers trained to predict the outputs of a hidden Markov model, for which residual activations are known to linearly encode a belief distribution over latent states in a probability simplex.

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。

↑