This theoretical study establishes a novel correspondence between signal propagation in deep neural networks and the absorbing state phase transition in statistical physics, providing valuable insights for a deeper understanding and systematic design of deep learning models.

Universal Scaling Laws of Absorbing Phase Transitions in Artificial Deep Neural Networks

Keiichi Tamai ¹, Tsuyoshi Okubo ¹, Truong Vinh Truong Duy ²,Naotake Natori ², Synge Todo ¹
1: The University of Tokyo
2: Aisin Corporation

関連リンク：
URL:https://journals.aps.org/prresearch/abstract/10.1103/jp61-6sp2
DOI: 10.1103/jp61-6sp2
Press release by the University of Tokyo: https://www.s.u-tokyo.ac.jp/en/press/10865/