Searcharxiv⌕ Search

arXiv subjects

Yoshiyuki Kabashima

Publications and source records attributed to Yoshiyuki Kabashima.

At least 109 records · Page 6Linked to original sources

Finite size scaling of the de Almeida-Thouless instability in random sparse networks

We study, in random sparse networks, finite size scaling of the spin glass susceptibility $χ_{\rm SG}$, which is a proper measure of the de Almeida-Thouless (AT) instability of spin glass systems. Using a phenomenological argument regarding the band edge behavior of the Hessian eigenvalue distribution, we discuss how $χ_{\rm SG}$ is evaluated in infinitely large random sparse networks, which are usually identified with Bethe trees, and how it should be corrected in finite systems. In the high temperature region, data of extensive numerical experiments are generally in good agreement with the theoretical values of $χ_{\rm SG}$ determined from the Bethe tree. In the absence of external fields, the data also show a scaling relation $χ_{\rm SG}=N^{1/3}F(N^{1/3}|T-T_c|/T_c)$, which has been conjectured in the literature, where $T_c$ is the critical temperature. In the presence of external fields, on the other hand, the numerical data are not consistent with this scaling relation. A numerical analysis of Hessian eigenvalues implies that strong finite size corrections of the lower band edge of the eigenvalue distribution, which seem relevant only in the presence of the fields, are a major source of inconsistency. This may be related to the known difficulty in using only numerical methods to detect the AT instability.

cond-mat.stat-mech↗

Zero-Temperature Complex Replica Zeros of the $\pm J$ Ising Spin Glass on Mean-Field Systems and Beyond

Zeros of the moment of the partition function $[Z^n]_{\bm{J}}$ with respect to complex $n$ are investigated in the zero temperature limit $β\to \infty$, $n\to 0$ keeping $y=βn \approx O(1)$. We numerically investigate the zeros of the $\pm J$ Ising spin glass models on several Cayley trees and hierarchical lattices and compare those results. In both lattices, the calculations are carried out with feasible computational costs by using recursion relations originated from the structures of those lattices. The results for Cayley trees show that a sequence of the zeros approaches the real axis of $y$ implying that a certain type of analyticity breaking actually occurs, although it is irrelevant for any known replica symmetry breaking. The result of hierarchical lattices also shows the presence of analyticity breaking, even in the two dimensional case in which there is no finite-temperature spin-glass transition, which implies the existence of the zero-temperature phase transition in the system. A notable tendency of hierarchical lattices is that the zeros spread in a wide region of the complex $y$ plane in comparison with the case of Cayley trees, which may reflect the difference between the mean-field and finite-dimensional systems.

cond-mat.dis-nn↗

Cavity approach to the first eigenvalue problem in a family of symmetric random sparse matrices

A methodology to analyze the properties of the first (largest) eigenvalue and its eigenvector is developed for large symmetric random sparse matrices utilizing the cavity method of statistical mechanics. Under a tree approximation, which is plausible for infinitely large systems, in conjunction with the introduction of a Lagrange multiplier for constraining the length of the eigenvector, the eigenvalue problem is reduced to a bunch of optimization problems of a quadratic function of a single variable, and the coefficients of the first and the second order terms of the functions act as cavity fields that are handled in cavity analysis. We show that the first eigenvalue is determined in such a way that the distribution of the cavity fields has a finite value for the second order moment with respect to the cavity fields of the first order coefficient. The validity and utility of the developed methodology are examined by applying it to two analytically solvable and one simple but non-trivial examples in conjunction with numerical justification.

math.OC↗

Weight space structure and analysis using a finite replica number in the Ising perceptron

The weight space of the Ising perceptron in which a set of random patterns is stored is examined using the generating function of the partition function $ϕ(n)=(1/N)\log [Z^n]$ as the dimension of the weight vector $N$ tends to infinity, where $Z$ is the partition function and $[ ... ]$ represents the configurational average. We utilize $ϕ(n)$ for two purposes, depending on the value of the ratio $α=M/N$, where $M$ is the number of random patterns. For $α< α_{\rm s}=0.833 ...$, we employ $ϕ(n)$, in conjunction with Parisi's one-step replica symmetry breaking scheme in the limit of $n \to 0$, to evaluate the complexity that characterizes the number of disjoint clusters of weights that are compatible with a given set of random patterns, which indicates that, in typical cases, the weight space is equally dominated by a single large cluster of exponentially many weights and exponentially many small clusters of a single weight. For $α> α_{\rm s}$, on the other hand, $ϕ(n)$ is used to assess the rate function of a small probability that a given set of random patterns is atypically separable by the Ising perceptrons. We show that the analyticity of the rate function changes at $α= α_{\rm GD}=1.245 ... $, which implies that the dominant configuration of the atypically separable patterns exhibits a phase transition at this critical ratio. Extensive numerical experiments are conducted to support the theoretical predictions.

cond-mat.dis-nn↗

Statistical mechanical analysis of the Kronecker channel model for MIMO wireless communication

The Kronecker channel model of wireless communication is analyzed using statistical mechanics methods. In the model, spatial proximities among transmission/reception antennas are taken into account as certain correlation matrices, which generally yield non-trivial dependence among symbols to be estimated. This prevents accurate assessment of the communication performance by naively using a previously developed analytical scheme based on a matrix integration formula. In order to resolve this difficulty, we develop a formalism that can formally handle the correlations in Kronecker models based on the known scheme. Unfortunately, direct application of the developed scheme is, in general, practically difficult. However, the formalism is still useful, indicating that the effect of the correlations generally increase after the fourth order with respect to correlation strength. Therefore, the known analytical scheme offers a good approximation in performance evaluation when the correlation strength is sufficiently small. For a class of specific correlation, we show that the performance analysis can be mapped to the problem of one-dimensional spin systems in random fields, which can be investigated without approximation by the belief propagation algorithm.

cs.IT↗

On analyticity with respect to the replica number in random energy models II: zeros on the complex plane

We characterize the breaking of analyticity with respect to the replica number which occurs in random energy models via the complex zeros of the moment of the partition function. We perturbatively evaluate the zeros in the vicinity of the transition point based on an exact expression of the moment of the partition function utilizing the steepest descent method, and examine an asymptotic form of the locus of the zeros as the system size tends to infinity. The incident angle of this locus indicates that analyticity breaking is analogous to a phase transition of the second order. We also evaluate the number of zeros utilizing the argument principle of complex analysis. The actual number of zeros calculated numerically for systems of finite size agrees fairly well with the analytical results.

cond-mat.dis-nn↗

On analyticity with respect to the replica number in random energy models I: an exact expression of the moment of the partition function

We provide an exact expression of the moment of the partition function for random energy models of finite system size, generalizing an earlier expression for a grand canonical version of the discrete random energy model presented by the authors in Prog. Theor. Phys. 111, 661 (2004). The expression can be handled both analytically and numerically, which is useful for examining how the analyticity of the moment with respect to the replica numbers, which play the role of powers of the moment, can be broken in the thermodynamic limit. A comparison with a replica method analysis indicates that the analyticity breaking can be regarded as the origin of the one-step replica symmetry breaking. The validity of the expression is also confirmed by numerical methods for finite systems.

cond-mat.dis-nn↗

Complex Replica Zeros of $\pm J$ Ising Spin Glass at Zero Temperature

Zeros of the $n$th moment of the partition function $[Z^n]$ are investigated in a vanishing temperature limit $β\to \infty$, $n \to 0$ keeping $y=βn \sim O(1)$. In this limit, the moment parameterized by $y$ characterizes the distribution of the ground-state energy. We numerically investigate the zeros for $\pm J$ Ising spin glass models with several ladder and tree systems, which can be carried out with a feasible computational cost by a symbolic operation based on the Bethe--Peierls method. For several tree systems we find that the zeros tend to approach the real axis of $y$ in the thermodynamic limit implying that the moment cannot be described by a single analytic function of $y$ as the system size tends to infinity, which may be associated with breaking of the replica symmetry. However, examination of the analytical properties of the moment function and assessment of the spin-glass susceptibility indicate that the breaking of analyticity is relevant to neither one-step or full replica symmetry breaking.

cond-mat.dis-nn↗

Learning of correlated patterns by simple perceptrons

Learning behavior of simple perceptrons is analyzed for a teacher-student scenario in which output labels are provided by a teacher network for a set of possibly correlated input patterns, and such that teacher and student networks are of the same type. Our main concern is the effect of statistical correlations among the input patterns on learning performance. For this purpose, we extend to the teacher-student scenario a methodology for analyzing randomly labeled patterns recently developed in {\em J. Phys. A: Math. Theor.} {\bf 41}, 324013 (2008). This methodology is used for analyzing situations in which orthogonality of the input patterns is enhanced in order to optimize the learning performance.

cond-mat.dis-nn↗

How could the replica method improve accuracy of performance assessment of channel coding?

We explore the relation between the techniques of statistical mechanics and information theory for assessing the performance of channel coding. We base our study on a framework developed by Gallager in {\em IEEE Trans. Inform. Theory} {\bf 11}, 3 (1965), where the minimum decoding error probability is upper-bounded by an average of a generalized Chernoff's bound over a code ensemble. We show that the resulting bound in the framework can be directly assessed by the replica method, which has been developed in statistical mechanics of disordered systems, whereas in Gallager's original methodology further replacement by another bound utilizing Jensen's inequality is necessary. Our approach associates a seemingly {\em ad hoc} restriction with respect to an adjustable parameter for optimizing the bound with a phase transition between two replica symmetric solutions, and can improve the accuracy of performance assessments of general code ensembles including low density parity check codes, although its mathematical justification is still open.

cs.IT↗

An integral formula for large random rectangular matrices and its application to analysis of linear vector channels

A statistical mechanical framework for analyzing random linear vector channels is presented in a large system limit. The framework is based on the assumptions that the left and right singular value bases of the rectangular channel matrix $\bH$ are generated independently from uniform distributions over Haar measures and the eigenvalues of $\bH^{\rm T}\bH$ asymptotically follow a certain specific distribution. These assumptions make it possible to characterize the communication performance of the channel utilizing an integral formula with respect to $\bH$, which is analogous to the one introduced by Marinari {\em et. al.} in {\em J. Phys. A} {\bf 27}, 7647 (1994) for large random square (symmetric) matrices. A computationally feasible algorithm for approximately decoding received signals based on the integral formula is also provided.

cs.IT↗

Inference from correlated patterns: a unified theory for perceptron learning and linear vector channels

A framework to analyze inference performance in densely connected single-layer feed-forward networks is developed for situations where a given data set is composed of correlated patterns. The framework is based on the assumption that the left and right singular value bases of the given pattern matrix are generated independently and uniformly from Haar measures. This assumption makes it possible to characterize the objective system by a single function of two variables which is determined by the eigenvalue spectrum of the cross-correlation matrix of the pattern matrix. Links to existing methods for analysis of perceptron learning and Gaussian linear vector channels and an application to a simple but nontrivial problem are also shown.

cs.IT↗

Perceptron capacity revisited: classification ability for correlated patterns

In this paper, we address the problem of how many randomly labeled patterns can be correctly classified by a single-layer perceptron when the patterns are correlated with each other. In order to solve this problem, two analytical schemes are developed based on the replica method and Thouless-Anderson-Palmer (TAP) approach by utilizing an integral formula concerning random rectangular matrices. The validity and relevance of the developed methodologies are shown for one known result and two example problems. A message-passing algorithm to perform the TAP scheme is also presented.

cond-mat.dis-nn↗

Statistical mechanical analysis of the linear vector channel in digital communication

A statistical mechanical framework to analyze linear vector channel models in digital wireless communication is proposed for a large system. The framework is a generalization of that proposed for code-division multiple-access systems in Europhys. Lett. 76 (2006) 1193 and enables the analysis of the system in which the elements of the channel transfer matrix are statistically correlated with each other. The significance of the proposed scheme is demonstrated by assessing the performance of an existing model of multi-input multi-output communication systems.

cs.IT↗

Analysis of CDMA systems that are characterized by eigenvalue spectrum

An approach by which to analyze the performance of the code division multiple access (CDMA) scheme, which is a core technology used in modern wireless communication systems, is provided. The approach characterizes the objective system by the eigenvalue spectrum of a cross-correlation matrix composed of signature sequences used in CDMA communication, which enables us to handle a wider class of CDMA systems beyond the basic model reported by Tanaka. The utility of the novel scheme is shown by analyzing a system in which the generation of signature sequences is designed for enhancing the orthogonality.

cs.IT↗

Numerical Study of TAP Metastable States in 3-body Ising Spin Glasses

The distribution of solutions of the Thouless-Anderson-Palmer equation is studied by extensive numerical experiments for fully connected 3-body interaction Ising spin glass models in a level of annealed calculation. A recent study predicted that when the equilibrium state of the system is characterized by one-step replica symmetry breaking, the distribution is described by a Becchi-Rouet-Stora-Tyutin (BRST) supersymmetric solution in the relatively low free energy region, whereas the BRST supersymmetry is broken for higher values of free energy (Crisanti et al., Phys. Rev. B 71 (2005) 094202). Our experiments qualitatively reproduce the discriminative behavior of macroscopic variables predicted by the theoretical assessment.

cond-mat.dis-nn↗

Survey propagation for the cascading Sourlas code

We investigate how insights from statistical physics, namely survey propagation, can improve decoding of a particular class of sparse error correcting codes. We show that a recently proposed algorithm, time averaged belief propagation, is in fact intimately linked to a specific survey propagation for which Parisi's replica symmetry breaking parameter is set to zero, and that the latter is always superior to belief propagation in the high connectivity limit. We briefly look at further improvements available by going to the second level of replica symmetry breaking.

cond-mat.stat-mech↗

Replicated Bethe Free Energy: A Variational Principle behind Survey Propagation

A scheme to provide various mean-field-type approximation algorithms is presented by employing the Bethe free energy formalism to a family of replicated systems in conjunction with analytical continuation with respect to the number of replicas. In the scheme, survey propagation (SP), which is an efficient algorithm developed recently for analyzing the microscopic properties of glassy states for a fixed sample of disordered systems, can be reproduced by assuming the simplest replica symmetry on stationary points of the replicated Bethe free energy. Belief propagation and generalized SP can also be offered in the identical framework under assumptions of the highest and broken replica symmetries, respectively.

cond-mat.dis-nn↗