Condensed Matter

Simplicial complexes are gaining increasing scientific attention as they are generalized network structures that can represent the many-body interactions existing in complex systems raging from the brain to high-order social networks. Simplicial complexes are formed by simplicies, such as nodes, links, triangles and so on. Cell complexes further extend these generalized network structures as they are formed by regular polytopes such as squares, pentagons etc. Pseudo-fractal simplicial and cell complexes are a major example of generalized network structures and they can be obtained by gluing 2 -dimensional m -polygons ( m=2 triangles, m=4 squares, m=5 pentagons, etc.) along their links according to a simple iterative rule. Here we investigate the interplay between the topology of pseudo-fractal simplicial and cell complexes and their dynamics by characterizing the critical properties of link percolation defined on these structures. By using the renormalization group we show that the pseudo-fractal simplicial and cell complexes have a continuous percolation threshold at p c =0 . When the pseudo-fractal structure is formed by polygons of the same size m , the transition is characterized by an exponential suppression of the order parameter P ∞ that depends on the number of sides m of the polygons forming the pseudo-fractal cell complex, i.e., P ∞ ∝pexp(−α/ p m−2 ) . Here these results are also generalized to random pseudo-fractal cell-complexes formed by polygons of different number of sides m .

We study the wave function localization properties in a d-dimensional model of randomly spaced particles with isotropic hopping potential depending solely on Euclidean interparticle distances. Due to the generality of this model usually called the Euclidean random matrix model, it arises naturally in various physical contexts such as studies of vibrational modes, artificial atomic systems, liquids and glasses, ultracold gases and photon localization phenomena. We generalize the known Burin-Levitov renormalization group approach, formulate universal conditions sufficient for localization in such models and inspect a striking equivalence of the wave function spatial decay between Euclidean random matrices and translation-invariant long-range lattice models with a diagonal disorder.

Nearly all statistical inference methods were developed for the regime where the number N of data samples is much larger than the data dimension p . Inference protocols such as maximum likelihood (ML) or maximum a posteriori probability (MAP) are unreliable if p=O(N) , due to overfitting. This limitation has for many disciplines with increasingly high-dimensional data become a serious bottleneck. We recently showed that in Cox regression for time-to-event data the overfitting errors are not just noise but take mostly the form of a bias, and how with the replica method from statistical physics once can model and predict this bias and the noise statistics. Here we extend our approach to arbitrary generalized linear regression models (GLM), with possibly correlated covariates. We analyse overfitting in ML/MAP inference without having to specify data types or regression models, relying only on the GLM form, and derive generic order parameter equations for the case of L2 priors. Second, we derive the probabilistic relationship between true and inferred regression coefficients in GLMs, and show that, for the relevant hyperparameter scaling and correlated covariates, the L2 regularization causes a predictable direction change of the coefficient vector. Our results, illustrated by application to linear, logistic, and Cox regression, enable one to correct ML and MAP inferences in GLMs systematically for overfitting bias, and thus extend their applicability into the hitherto forbidden regime p=O(N) .

In their comment, Angelini et al. object to the conclusion of [J. Phys. A: Math. Theor., 52:445002, 2019] (1), where we show that in [Phys. Rev. B, 87:134201, 2013] the exponent ν has been obtained by applying a mathematical relation in a regime where this relation is not valid. We observe that the criticism above on the mathematical validity of such relation has not been addressed in the comment. Our criticism thus remains valid, and disproves the conclusions of the comment. This constitutes the main point of this reply. We also provide a point-by-point response and discussion of Angelini et al.'s claims. First, Angelini et al. claim that the prediction 2 1/ν =1 of [1] is incorrect, because it results from the relation λ max = 2 1/ν between the largest eigenvalue of the linearized renormalization-group (RG) transformation and ν , which cannot be applied to the ensemble renormalization group (ERG) method, because for the ERG λ max =1 . However, the feature λ max =1 is specific to the ERG transformation and it does not give any grounds for questioning the validity of the general relation λ max = 2 1/ν specifically for the ERG transformation. Second, Angelini et al. claim that ν should be extracted from an early RG regime (A), as opposed to the asymptotic regime (B) used to estimate ν in [1] and that (B) is dominated by finite-size effects. Still, (A) is a small-wavelength, non-critical regime, which cannot characterize the critical exponent ν related to the divergence of the correlation length. Also, the fact that (B) involves finite-size effects is a feature specific to the ERG, and gives no rationale for extracting ν from (A). Finally, we refute the remaining claims made by Angelini et al., and thus stand by our assertion that the ERG method yields a prediction given by 2 1/ν =1 .

The topology of a network associated with a reservoir computer is often taken so that the connectivity and the weights are chosen randomly. Optimization is hardly considered as the parameter space is typically too large. Here we investigate this problem for a class of reservoir computers for which we obtain a decomposition of the reservoir dynamics into modes, which can be computed independently of one another. Each mode depends on an eigenvalue of the network adjacency matrix. We then take a parametric approach in which the eigenvalues are parameters that can be appropriately designed and optimized. In addition, we introduce the application of a time shift to each individual mode. We show that manipulations of the individual modes, either in terms of the eigenvalues or the time shifts, can lead to dramatic reductions in the training error.

This review deals with Restricted Boltzmann Machine (RBM) under the light of statistical physics. The RBM is a classical family of Machine learning (ML) models which played a central role in the development of deep learning. Viewing it as a Spin Glass model and exhibiting various links with other models of statistical physics, we gather recent results dealing with mean-field theory in this context. First the functioning of the RBM can be analyzed via the phase diagrams obtained for various statistical ensembles of RBM leading in particular to identify a {\it compositional phase} where a small number of features or modes are combined to form complex patterns. Then we discuss recent works either able to devise mean-field based learning algorithms; either able to reproduce generic aspects of the learning process from some {\it ensemble dynamics equations} or/and from linear stability arguments.

In this work, the capability of restricted Boltzmann machines (RBMs) to find solutions for the Kitaev honeycomb model with periodic boundary conditions is investigated. The measured groundstate (GS) energy of the system is compared and, for small lattice sizes (e.g. 3×3 with 18 spinors), shown to agree with the analytically derived value of the energy up to a deviation of 0.09% . Moreover, the wave-functions we find have 99.89% overlap with the exact ground state wave-functions. Furthermore, the possibility of realizing anyons in the RBM is discussed and an algorithm is given to build these anyonic excitations and braid them for possible future applications in quantum computation. Using the correspondence between topological field theories in (2+1)d and 2d CFTs, we propose an identification between our RBM states with the Moore-Read state and conformal blocks of the 2 d Ising model.

We study the first-order flocking transition of birds flying in low-visibility conditions by employing three different representative types of neural network (NN) based machine learning architectures that are trained via either an unsupervised learning approach called "learning by confusion" or a widely used supervised learning approach. We find that after the training via either the unsupervised learning approach or the supervised learning one, all of these three different representative types of NNs, namely, the fully-connected NN, the convolutional NN, and the residual NN, are able to successfully identify the first-order flocking transition point of this nonequilibrium many-body system. This indicates that NN based machine learning can be employed as a promising generic tool to investigate rich physics in scenarios associated to first-order phase transitions and nonequilibrium many-body systems.

We report a metadynamics simulation study of crystallization in a deep undercooled metallic glass-forming liquid by developing appropriate collective variables. Through a combined analysis of free energy surface (FES) and atomic-level behaviors, a picture of an abnormal-endothermic crystallization process is revealed: rejuvenated disorder states with less local fivefold-symmetry and fast dynamics form firstly by changing the local chemical order around Cu atoms, which then act as the precursor for the nucleation of well-ordered crystallites. This process reflects a complex energy landscape with well-separated glassy and crystal basins, giving rise to the direct evidence of intrinsic frustration against crystallization in deep undercooled metallic glass forming liquids. Moreover, the rejuvenated disorder states with distinct physical behaviors offer great opportunities to tailor the performances of metallic glass by controlling the thermal history of a metallic melt.

Disordered systems like liquids, gels, glasses, or granular materials are not only ubiquitous in daily life and in industrial applications but they are also crucial for the mechanical stability of cells or the transport of chemical and biological agents in living organisms. Despite the importance of these systems, their microscopic structure is understood only on a rudimentary level, thus in stark contrast to the case of gases and crystals. Since scattering experiments and analytical calculations usually give only structural information that is spherically averaged, the three dimensional (3D) structure of disordered systems is basically unknown. Here we introduce a simple method that allows to probe the 3D structure of such systems. Using computer simulations we find that hard-sphere-like liquids have on intermediate and large scales an intricate structural order given by alternating layers with icosahedral and dodecahedral symmetries, while open network liquids like silica have a structural order with tetrahedral symmetry. These results show that liquids have a highly non-trivial 3D structure and that this structural information is encoded in non-standard correlation functions.