arXiv subjects

Leonid Bogachev

Publications and source records attributed to Leonid Bogachev.

7 recordsLinked to original sources

arXiv2026

Can LLM Reasoning Be Trusted? A Comparative Study: Using Human Benchmarking on Statistical Tasks

This paper investigates the ability of large language models (LLMs) to solve statistical tasks, as well as their capacity to assess the quality of reasoning. While state-of-the-art LLMs have demonstrated remarkable performance in a range of NLP tasks, their competence in addressing even moderately complex statistical challenges is not well understood. We have fine-tuned selected open-source LLMs on a specially developed dataset to enhance their statistical reasoning capabilities, and compared their performance with the human scores used as a benchmark. Our results show that the fine-tuned models achieve better performance on advanced statistical tasks on the level comparable to a statistics student. Fine-tuning demonstrates architecture-dependent improvements, with some models showing significant performance gains, indicating clear potential for deployment in educational technology and statistical analysis assistance systems. We also show that LLMs themselves can be far better judges of the answers quality (including explanation and reasoning assessment) in comparison to traditional metrics, such as BLEU or BertScore. This self-evaluation capability enables scalable automated assessment for statistical education platforms and quality assurance in automated analysis tools. Potential applications also include validation tools for research methodology in academic and industry settings, and quality control mechanisms for data analysis workflows.

cs.CL↗

arXiv2023

The joint node degree distribution in the Erdős-Rényi network

The Erdős-Rényi random graph is the simplest model for node degree distribution, and it is one of the most widely studied. In this model, pairs of $n$ vertices are selected and connected uniformly at random with probability $p$, consequently, the degrees for a given vertex follow the binomial distribution. If the number of vertices is large, the binomial can be approximated by Normal using the Central Limit Theorem, which is often allowed when $\min (np, n(1-p)) > 5$. This is true for every node independently. However, due to the fact that the degrees of nodes in a graph are not independent, we aim in this paper to test whether the degrees of per node collectively in the Erdős-Rényi graph have a multivariate normal distribution MVN. A chi square goodness of fit test for the hypothesis that binomial is a distribution for the whole set of nodes is rejected because of the dependence between degrees. Before testing MVN we show that the covariance and correlation between the degrees of any pair of nodes in the graph are $p(1-p)$ and $1/(n-1)$, respectively. We test MVN considering two assumptions: independent and dependent degrees, and we obtain our results based on the percentages of rejected statistics of chi square, the $p$-values of Anderson Darling test, and a CDF comparison. We always achieve a good fit of multivariate normal distribution with large values of $n$ and $p$, and very poor fit when $n$ or $p$ are very small. The approximation seems valid when $np \geq 10$. We also compare the maximum likelihood estimate of $p$ in MVN distribution where we assume independence and dependence. The estimators are assessed using bias, variance and mean square error.

stat.ML↗

arXiv2011

Cluster point processes on manifolds

The probability distribution $μ_{cl}$ of a general cluster point process in a Riemannian manifold $X$ (with independent random clusters attached to points of a configuration with distribution $μ$) is studied via the projection of an auxiliary measure $\hatμ$ in the space of configurations $\hatγ=\{(x,\bar{y})\}\subset X\times\mathfrak{X}$, where $x\in X$ indicates a cluster "centre" and $\bar{y}\in\mathfrak{X}:=\bigsqcup_{n} X^n$ represents a corresponding cluster relative to $x$. We show that the measure $μ_{cl}$ is quasi-invariant with respect to the group $Diff_{0}(X)$ of compactly supported diffeomorphisms of $X$, and prove an integration-by-parts formula for $μ_{cl}$. The associated equilibrium stochastic dynamics is then constructed using the method of Dirichlet forms. General constructions are illustrated by examples including Euclidean spaces, Lie groups, homogeneous spaces, Riemannian manifolds of nonpositive curvature and metric spaces. The paper is an extension of our earlier results for Poisson cluster measures [J. Funct. Analysis 256 (2009) 432-478] and for Gibbs cluster measures [arxiv:1007.3148], where different projection constructions were utilised.

math.FA↗

arXiv2010

Occupation time distributions for the telegraph process

For the one-dimensional telegraph process, we obtain explicit distribution of the occupation time of the positive half-line. The long-term limiting distribution is then derived when the initial location of the process is in the range of sub-normal or normal deviations from the origin; in the former case, the limit is given by the arcsine law. These limit theorems are also extended to the case of more general occupation-type functionals.

math.PR↗

arXiv2010

Gibbs cluster measures on configuration spaces

The distribution $g_{cl}$ of a Gibbs cluster point process in $X=\mathbb{R}^{d}$ (with i.i.d. random clusters attached to points of a Gibbs configuration with distribution $g$) is studied via the projection of an auxiliary Gibbs measure $\hat{g}$ in the space of configurations $\hat{gamma}=\{(x,\bar{y})\}\subset X\times\mathfrak{X}$, where $x\in X$ indicates a cluster "center" and $\bar{y}\in\mathfrak{X}:=\bigsqcup_{n} X^n$ represents a corresponding cluster relative to $x$. We show that the measure $g_{cl}$ is quasi-invariant with respect to the group $\mathrm{Diff}_{0}(X)$ of compactly supported diffeomorphisms of $X$, and prove an integration-by-parts formula for $g_{cl}$. The associated equilibrium stochastic dynamics is then constructed using the method of Dirichlet forms. These results are quite general; in particular, the uniqueness of the background Gibbs measure $g$ is not required. The paper is an extension of the earlier results for Poisson cluster measures %obtained by the authors [J. Funct. Analysis 256 (2009) 432-478], where a different projection construction was utilized specific to this "exactly soluble" case.

math.FA↗

arXiv2008

Poisson Cluster Measures: Quasi-invariance, Integration by Parts and Equilibrium Stochastic Dynamics

The distribution $μ_{cl}$ of a Poisson cluster process in $X=\mathbb{R}^{d}$ (with i.i.d. clusters) is studied via an auxiliary Poisson measure on the space of configurations in $\mathfrak{X}=\sqcup_{n} X^n$, with intensity measure defined as a convolution of the background intensity of cluster centres and the probability distribution of a generic cluster. We show that the measure $μ_{cl}$ is quasi-invariant with respect to the group of compactly supported diffeomorphisms of $X$ and prove an integration-by-parts formula for $μ_{cl}$. The corresponding equilibrium stochastic dynamics is then constructed using the method of Dirichlet forms.

math.FA↗

arXiv2007

On bounded solutions of the balanced generalized pantograph equation

The question about the existence and characterization of bounded solutions to linear functional-differential equations with both advanced and delayed arguments was posed in early 1970s by T. Kato in connection with the analysis of the pantograph equation, y'(x)=ay(qx)+by(x). In the present paper, we answer this question for the balanced generalized pantograph equation of the form -a_2 y''(x)+a_1 y'(x)+y(x)=int_0^infty y(qx) m(dq), where a_1 > or = 0, a_2 > or = 0, a_1^2+a_2^2>0, and m is a probability measure. Namely, setting K:=int_0^infty ln(q) m(dq), we prove that if K < or = 0 then the equation does not have nontrivial (i.e., nonconstant) bounded solutions, while if K>0 then such a solution exists. The result in the critical case, K=0, settles a long-standing problem. The proof exploits the link with the theory of Markov processes, in that any solution of the balanced pantograph equation is an L-harmonic function relative to the generator L of a certain diffusion process with "multiplication" jumps. The paper also includes three "elementary" proofs for the simple prototype equation y'(x)+y(x)=(1/2)y(qx)+(1/2)y(x/q), based on perturbation, analytical, and probabilistic techniques, respectively, which may appear useful in other situations as efficient exploratory tools.

math.PR↗