-
Absorbing Phase Transitions in Artificial Deep Neural Networks
Authors:
Keiichi Tamai,
Tsuyoshi Okubo,
Truong Vinh Truong Duy,
Naotake Natori,
Synge Todo
Abstract:
Theoretical understanding of the behavior of infinitely-wide neural networks has been rapidly developed for various architectures due to the celebrated mean-field theory. However, there is a lack of a clear, intuitive framework for extending our understanding to finite networks that are of more practical and realistic importance. In the present contribution, we demonstrate that the behavior of pro…
▽ More
Theoretical understanding of the behavior of infinitely-wide neural networks has been rapidly developed for various architectures due to the celebrated mean-field theory. However, there is a lack of a clear, intuitive framework for extending our understanding to finite networks that are of more practical and realistic importance. In the present contribution, we demonstrate that the behavior of properly initialized neural networks can be understood in terms of universal critical phenomena in absorbing phase transitions. More specifically, we study the order-to-chaos transition in the fully-connected feedforward neural networks and the convolutional ones to show that (i) there is a well-defined transition from the ordered state to the chaotics state even for the finite networks, and (ii) difference in architecture is reflected in that of the universality class of the transition. Remarkably, the finite-size scaling can also be successfully applied, indicating that intuitive phenomenological argument could lead us to semi-quantitative description of the signal propagation dynamics.
△ Less
Submitted 5 July, 2023;
originally announced July 2023.
-
Data Assimilation Method for Experimental and First-Principles Data: Finite-Temperature Magnetization of (Nd,Pr,La,Ce)$_{2}$(Fe,Co,Ni)$_{14}$B
Authors:
Yosuke Harashima,
Keiichi Tamai,
Shotaro Doi,
Munehisa Matsumoto,
Hisazumi Akai,
Naoki Kawashima,
Masaaki Ito,
Noritsugu Sakuma,
Akira Kato,
Tetsuya Shoji,
Takashi Miyake
Abstract:
We propose a data-assimilation method for evaluating the finite-temperature magnetization of a permanent magnet over a high-dimensional composition space. Based on a general framework for constructing a predictor from two data sets including missing values, a practical scheme for magnetic materials is formulated in which a small number of experimental data in limited composition space are integrat…
▽ More
We propose a data-assimilation method for evaluating the finite-temperature magnetization of a permanent magnet over a high-dimensional composition space. Based on a general framework for constructing a predictor from two data sets including missing values, a practical scheme for magnetic materials is formulated in which a small number of experimental data in limited composition space are integrated with a larger number of first-principles calculation data. We apply the scheme to (Nd$_{1-α-β-γ}$Pr$_α$La$_β$Ce$_γ$)$_{2}$(Fe$_{1-δ-ζ}$Co$_δ$Ni$_ζ$)$_{14}$B. The magnetization in the whole $(α, β, γ, δ, ζ)$ space at arbitrary temperature is obtained. It is shown that the Co doping does not enhance the magnetization at low temperatures, whereas the magnetization increases with increasing $δ$ above 320 K.
△ Less
Submitted 28 July, 2020;
originally announced July 2020.
-
How to experimentally probe universal features of absorbing phase transitions using steady state
Authors:
Keiichi Tamai,
Masaki Sano
Abstract:
We propose experimentally feasible ways to probe universal features of absorbing phase transitions from two different approaches, both based on numerical validations. On one hand, we numerically study a probability distribution of duration/length of intervals of local inactive state in quasi-steady state, which has been very commonly used in experiments, in a case of the contact process. We show t…
▽ More
We propose experimentally feasible ways to probe universal features of absorbing phase transitions from two different approaches, both based on numerical validations. On one hand, we numerically study a probability distribution of duration/length of intervals of local inactive state in quasi-steady state, which has been very commonly used in experiments, in a case of the contact process. We show that the distributions obey the universal scaling ansatz expected from phenomenological scaling argument, but that care must be taken in order to suppress a bias caused by censoring due to a finite observation window. To demonstrate the latter point, we compare the distributions for the temporal intervals estimated through conventional histograms with those through the estimator which properly takes account of censoring and sampling bias. On the other hand, we also propose that, if a system is subject to uniform advection as is often the case with flowing systems, a correlation length and a correlation time near the transition point can be easily quantified by supplying the system with an active boundary condition. In order to support our proposal, we introduce a new model whose advection strength can be arbitrarily controlled. The results of numerical simulations on our model suggest that a correlation time, which is difficult to measure through the interval distributions without the aforementioned bias, can be measured through characteristic decay length of an order parameter. Crossovers between two different power-law behaviors are also identified in this case, and the universal scaling ansätze associated with the crossovers are discussed.
△ Less
Submitted 23 October, 2018; v1 submitted 15 December, 2017;
originally announced December 2017.
-
A Universal Transition to Turbulence in Channel Flow
Authors:
Masaki Sano,
Keiichi Tamai
Abstract:
Transition from laminar to turbulent flow drastically changes the mixing, transport, and drag properties of fluids, yet when and how turbulence emerges is elusive even for simple flow within pipes and rectangular channels. Unlike the onset of temporal disorder, which is identified as the universal route to chaos in confined flows, characterization of the onset of spatio-temporal disorder has been…
▽ More
Transition from laminar to turbulent flow drastically changes the mixing, transport, and drag properties of fluids, yet when and how turbulence emerges is elusive even for simple flow within pipes and rectangular channels. Unlike the onset of temporal disorder, which is identified as the universal route to chaos in confined flows, characterization of the onset of spatio-temporal disorder has been an outstanding challenge because turbulent domains irregularly decay or spread as they propagate downstream. Here, through extensive experimental investigation of channel flow, we identify a distinctive transition with critical behavior. Turbulent domains continuously injected from an inlet ultimately decayed, or in contrast, spread depending on flow rates. Near a transition point, critical behavior was observed. We investigate both spatial and temporal dynamics of turbulent clusters, measuring four critical exponents, a universal scaling function and a scaling relation, all in agreement with the (2+1)-dimensional directed percolation universality class.
△ Less
Submitted 27 October, 2015;
originally announced October 2015.
