arXiv subjects

Matthias Löwe

Publications and source records attributed to Matthias Löwe.

At least 19 recordsLinked to original sources

arXiv2026

On associative neural networks for sparse patterns with huge capacities

Generalized Hopfield models with higher-order or exponential interaction terms are known to have substantially larger storage capacities than the classical quadratic model. On the other hand, associative memories for sparse patterns, such as the Willshaw and Amari models, already exhibit enhanced storage capacities in the sparse regime. In this paper we combine these two mechanisms. We introduce higher-order versions of sparse associative memory models and study their storage capacities in the sense of fixed-pattern stability. For the Amari and Willshaw models with fixed interaction order $n$, we obtain storage scales of order $\frac{N^n}{(\log N)^n}$. When the interaction order grows logarithmically with the number of neurons, the resulting storage scale becomes super-polynomial. We also study higher-order interactions in the block-structured Gripon--Berrou architecture, where the natural storage scale is of order $c^n$. Our results show that the capacity increase caused by higher-order interactions persists in the sparse setting, while the precise storage scale depends on the underlying architecture.

math.PR↗

arXiv2026

When does propagation of chaos in the critical Curie-Weiss model stop?

We study increasing propagation of chaos for the critical Curie- Weiss model (i.e. the mean-field Ising model at critical inverse temperature 1, with no external field). We give a simple way to see that for windows of size k(N) of smaller order than sqrt{N} we still have propagation of chaos (reproving earlier results, see e.g.[1]), while for k(N) of order sqrt{N} the propagation of chaos breaks down. The law of a single spin converges to π, the Bernoulli law with parameter 1/2. If k(N) = alpha sqrt{N}, we give an explicit formula for the limiting distance in total variation of the law of the first k(N) spins with respect to the k-fold product of pi, as a function of alpha. For even larger window sizes, the distribution of the spins has, in the thermodynamical limit, maximal distance to the k-fold product of pi. One of the ingredients of the proof is a result about the unimodality/non- unimodality of the law of the number of positive spins among the first k = k(N) spins, which may be of independent interest.

math.PR↗

arXiv2026

On the Effect of Bottlenecks in Block Spin Models

We study a bottleneck spin model with $N$ spins, split into two Curie-Weiss models at low temperature with a bottleneck between them. We propose multiple ways of how to realize such a bottleneck and study its influence on the phase transition in the thermodynamic limit $N \to \infty$. In all versions of this model we prove the existence of a threshold that determines whether or not the presence of the bottleneck is felt in the phase transition. This threshold depends on the size of the bottleneck and the interaction strength through it.

math.PR↗

arXiv2026

A mathematical analysis of hierarchical Hopfield models

The central question that we address is: How can structured information be stored in a hierarchical Hopfield model involving hidden layers? To this end, we develop a formalism of strokes and concepts that allows us to appropriately structure information: initial features are first classified into strokes, which in a second step are aggregated into concepts. We rigorously derive criteria under which concepts can be retrieved from noisy input data. A remarkable effect is that we do not require a perfect retrieval at the level of strokes, as the second-layer retrieval procedure compensates for first-layer errors. We treat separately the cases of fixed and variable-sized concepts.

math.PR↗

arXiv2026

The infinite block spin Ising model

We study a block mean-field Ising model with $N$ spins split into $s_N$ blocks, with Curie-Weiss interaction within blocks and nearest-neighbor coupling between blocks. While previous models deal with the block magnetization for a fixed number of blocks, we study the the simultaneous limit $N\to\infty$ and $s_N\to\infty$. The model interpolates between Curie-Weiss model for $s_N=1$, multi-species mean field for fixed $s_N=s$, and the 1D Ising model for each spin in its own block at $s_N=N$. Under mild growth conditions on $s_N$, we prove a law of large numbers and a multivariate CLT with covariance given by the lattice Green's function. For instance, the high temperature CLT essentially covers the optimal range up to $s_N=o(N/(\log N)^c)$ and the low temperature regime is new even for fixed number of blocks $s > 2$. In addition to the standard competition between entropy and energy, a new obstacle in the proofs is a curse of dimensionality as $s_N \to \infty$.

math.PR↗

arXiv2026

High-Temperature Increasing Propagation of Chaos and its breakdown for the Hopfield Model

We analyze increasing propagation of chaos in the high temperature regime of a disordered mean-field model, the Hopfield model. We show that for $β<1$ (the true high temperature region) we have increasing propagation of chaos as long as the size of the marginals $k=k(N)$ and the number of patterns $M=M(N)$ satisfies $Mk/N \to 0$. For $M=o(\sqrt N)$ we show that propagation of chaos breaks down for $k/N \to c>0$. At the ciritcal temperature we show that, for $M$ finite, there is increasing propagation of chaos, for $k=o(\sqrt N)$, while we have breakdown of propagation of chaos for $k=c \sqrt N$, for a $c>0$. All these reulst hold in probability in the disorder.

math.PR↗

arXiv2025

Spectral properties of the stochastic block model and their application to hitting times of random walks

We analyze hitting times of simple random walk on realizations of the stochastic block model. We show that under some natural assumptions the hitting time averaged over the target vertex asymptotically almost surely given by $N(1+o(1))$. On the other hand, the hitting time averaged over the starting vertex asymptotically almost surely depends on expected degrees in the block the target vertex is in. We also show a central limit theorem for the hitting time averaged over the starting vertex. Our main techniques are a spectral decomposition of these hitting times, a spectral analysis of the adjacency matrix and the graph Laplacian.

math.PR↗

arXiv2024

Propagation of chaos and residual dependence in Gibbs measures on finite sets

We compare a mean-field Gibbs distribution on a finite state space on $N$ spins to that of an explicit simple mixture of product measures. This illustrates the situation beyond the so-called increasing propagation of chaos introduced by Ben Arous and Zeitouni 1999, where marginal distributions of size $k=o(N)$ are compared to product measures.

math.PR↗

arXiv2023

Propagation of chaos in the Random field Curie-Weiss model

We prove propagation of chaos in the Random field mean-field Ising model, also known ad the Random field Curie-Weiss model. We show that in the paramagnetic phase, i.e.\ in the regime where temperature and distribution of the external field admit a unique minimizer of the expected Helmholtz free energy, propagation of chaos holds. By the latter we mean that the finite-dimensional marginals of the Gibbs measure converge towards a product measure with the correct expectation as the system size goes to infinity. This holds independently of whether the system is in a high-temperature phase or at a phase transition point. If the Helmholtz free energy possesses several minima, there are several possible equilibrium measures. In this case, we show that the system picks one of them at random (depending on the realization of the random external field) and propagation of chaos with respect to a product measure with the same marginals as the one randomly picked holds true. We illustrate our findings in a simple example.

math.PR↗

arXiv2023

A Central Limit Theorem for the average target hitting time for a random walk on a random graph

Consider a simple random walk on a realization of an Erdős-Rényi graph. Assume that it is asymptotically almost surely (a.a.s.) connected. Conditional on an eigenvector delocalization conjecture, we prove a Central Limit Theorem (CLT) for the average target hitting time. By the latter we mean the expected time it takes the random walk on average to first hit a vertex $j$ when starting in a fixed vertex $i$. The average is taken with respect to $π_i$, the invariant measure of the random walk.

math.PR↗

arXiv2023

When does the chaos in the Curie-Weiss model stop to propagate?

We investigate increasing propagation of chaos for the mean-field Ising model of ferromagnetism (also known as the Curie-Weiss model) with $N$ spins at inverse temperature $β>0$ and subject to an external magnetic field of strength $h\in\mathbb{R}$. Using a different proof technique than in [Ben Arous, Zeitouni; 1999] we confirm the well-known propagation of chaos phenomenon: If $k=k(N)=o(N)$ as $N\to\infty$, then the $k$'th marginal distribution of the Gibbs measure converges to a product measure at $β<1$ or $h \neq 0$ and to a mixture of two product measures, if $β>1$ and $h =0$. More importantly, we also show that if $k(N)/N\to α\in (0,1]$, this property is lost and we identify a non-zero limit of the total variation distance between the number of positive spins among any $k$-tuple and the corresponding binomial distribution.

math.PR↗

arXiv2022

Collaborative Insurance Sustainability and Network Structure

The peer-to-peer (P2P) economy has been growing with the advent of the Internet, with well known brands such as Uber or Airbnb being examples thereof. In the insurance sector the approach is still in its infancy, but some companies have started to explore P2P-based collaborative insurance products (eg. Lemonade in the U.S. or Inspeer in France). The actuarial literature only recently started to consider those risk sharing mechanisms, as in Denuit and Robert (2021) or Feng et al. (2021). In this paper, describe and analyse such a P2P product, with some reciprocal risk sharing contracts. Here, we consider the case where policyholders still have an insurance contract, but the first self-insurance layer, below the deductible, can be shared with friends. We study the impact of the shape of the network (through the distribution of degrees) on the risk reduction. We consider also some optimal setting of the reciprocal commitments, and discuss the introduction of contracts with friends of friends to mitigate some possible drawbacks of having people without enough connections to exchange risks.

q-fin.RM↗

arXiv2021

Reconstructing a (recurrent) random environment from a single trajectory of Random Walk in Random Environment with errors

We consider one infinite path of a Random Walk in Random Environment (RWRE, for short) in an unknown environment. This environment consists of either i.i.d.\ site or bond randomness. At each position the random walker stops and tells us the environment it sees at the point where it is, without telling us, where it is. These observations $χ'$ are spoiled by reading errors that occur with probability $p<1$. We show: If the RWRE is recurrent and satisfies the standard assumptions on such RWREs, then with probability one in the environment, the errors, and the random walk we are able reconstruct the law of the environment. For most situations this result is even independent of the value of $p$. If the distribution of the environment has a non-atomic part, we can even reconstruct the environment itself, up to translation.

math.PR↗

arXiv2021

Fluctuations of the magnetization in the Block Potts Model

In this note we study the block spin mean-field Potts model, in which the spins are divided into $s$ blocks and can take $q\ge 2$ different values (colors). Each block is allowed to contain a different proportion of vertices and behaves itself like a mean-field Ising/Potts model which also interacts with other blocks according to different temperatures. Of particular interest is the behavior of the magnetization, which counts the number of colors appearing in the distinct blocks. We prove central limit theorems for the magnetization in the generalized high temperature regime and provide a moderate deviation principle for its fluctuations on lower scalings. More precisely, the magnetization concentrates around the uniform vector of all colors with an explicit, but singular, Gaussian distribution. In order to remove the singular component, we will also consider a rotated magnetization, which enables us to compare our results to various related models.

math.PR↗

arXiv2021

Fluctuations of the Magnetization for Ising models on Erdős-Rényi Random Graphs -- the Regimes of Low Temperature and External Magnetic Field

We continue our analysis of Ising models on the (directed) Erdős-Rényi random graph $G(N,p)$. We prove a quenched Central Limit Theorem for the magnetization and describe the fluctuations of the log-partition function. In the current note we consider the low temperature regime $β>1$ and the case when an external magnetic field is present. In both cases, we assume that $p=p(N)$ satisfies $p^3N \to \infty$.

math.PR↗

arXiv2020

Some Remarks on Replicated Simulated Annealing

Recently authors have introduced the idea of training discrete weights neural networks using a mix between classical simulated annealing and a replica ansatz known from the statistical physics literature. Among other points, they claim their method is able to find robust configurations. In this paper, we analyze this so-called "replicated simulated annealing" algorithm. In particular, we explicit criteria to guarantee its convergence, and study when it successfully samples from configurations. We also perform experiments using synthetic and real data bases.

cs.LG↗

arXiv2020

Large deviations and a phase transition in the Block Spin Potts models

We introduce and analyze a generalization of the blocks spin Ising (Curie-Weiss) models that were discussed in a number of recent articles. In these block spin models each spin in one of $s$ blocks can take one of a finite number of $q \ge 3$ values, hence the name block spin Potts model. The values a spin can take are called colors. We prove a large deviation principle for the percentage of spins of a certain color in a certain block. These values are represented in an $s \times q$ matrix. We show that for uniform block sizes and appropriately chosen interaction strength there is a phase transition. In some regime the only equilibrium is the uniform distribution of all colors in all blocks, while in other parameter regimes there is one predominant color, and this is the same color with the same frequency for all blocks. Finally, we establish log-Sobolev-type inequalities for the block spin Potts model.

math.PR↗

arXiv2020

Towards an Intrinsic Definition of Robustness for a Classifier

The robustness of classifiers has become a question of paramount importance in the past few years. Indeed, it has been shown that state-of-the-art deep learning architectures can easily be fooled with imperceptible changes to their inputs. Therefore, finding good measures of robustness of a trained classifier is a key issue in the field. In this paper, we point out that averaging the radius of robustness of samples in a validation set is a statistically weak measure. We propose instead to weight the importance of samples depending on their difficulty. We motivate the proposed score by a theoretical case study using logistic regression, where we show that the proposed score is independent of the choice of the samples it is evaluated upon. We also empirically demonstrate the ability of the proposed score to measure robustness of classifiers with little dependence on the choice of samples in more complex settings, including deep convolutional neural networks and real datasets.

cs.CV↗