arXiv subjects

Zhiyi Chi

Publications and source records attributed to Zhiyi Chi.

At least 19 recordsLinked to original sources

arXiv2026

On difference in word frequencies in a symmetric Bernoulli process

In a symmetric Bernoulli process, all binary strings, or ``words'' of the same length have the same long term frequency. However, between two such words, one may have a ``frequency advantage'' in the sense that in any long enough segment of the Bernoulli process, the probability that the word occurs more times than the other word is greater than the probability the other way around. To characterize the frequency advantage in the long run, the asymptotics of the difference between the two probabilities as the length of the segment of the Bernoulli process tends to infinity is derived.

math.PR↗

arXiv2026

A Time-Varying Branching Process Approach to Model Self-Renewing Cells

Stem cells, through their ability to produce daughter stem cells and differentiate into specialized cells, are essential in the growth, maintenance, and repair of biological tissues. Understanding the dynamics of cell populations in the proliferation process not only uncovers proliferative properties of stem cells, but also offers insight into tissue development under both normal conditions and pathological disruption. In this paper, we develop a continuous time branching process model with time-dependent offspring distribution to characterize stem cell proliferation process. We derive analytical expressions for mean, variance, and autocovariance of the stem cell counts, and develop likelihood-based inference procedures to estimate model parameters. Particularly, we construct a forward algorithm likelihood to handle situations when some cell types cannot be directly observed. Simulation results demonstrate that our estimation method recovers the time-dependent division probabilities with good accuracy.

stat.AP↗

arXiv2025

Complexity of exact sampling of the first passage of a stable subordinator

We consider the exact sampling of the first passage of a stable subordinator across a non-increasing regular barrier. First, the sampling is reduced to one from a bivariate distribution parameterized by the index $\alpha$ of the subordinator and a scalar $z$ independent of the barrier. Then three algorithms are devised for different regions of $(\alpha, z)$, using the acceptance-rejection method without numerical inversion or integration. When combined, the algorithms allow the exact sampling of the first passage to be done with complexity $O(1+|\ln(1-\alpha)|)$.

stat.CO↗

arXiv2025

On a Variation of Gambler's Ruin Problem

Assume that letters (from a finite alphabet) in a text form a Markov chain. We track two distinct words, $U$ and $D$. A gambler gains 1 point for each occurrence of $U$ (including overlapping occurrences) and loses 1 point for each occurrence of $D$ (also including overlapping occurrences). We determine the probability of gaining $A$ points before losing $B$ points, where $A$ and $B$ are integers. Additionally, we find the expected waiting time until one of the two events -- gaining $A$ points or losing $B$ points -- occurs.

math.PR↗

arXiv2024

Recursive Computation of Path Homology for Stratified Digraphs

Stratified digraphs are popular models for feedforward neural networks. However, computation of their path homologies has been limited to low dimensional ones due to high computational complexity. A recursive algorithm is proposed to compute certain high-dimensional (reduced) path homologies of stratified digraphs. By recursion on matrix representations of homologies of subgraphs, the algorithm efficiently computes the full-depth path homology of a stratified digraph, i.e. homology with dimension equal to the depth of the graph. The algorithm can be used to compute the maximal path homology of acyclic digraphs, i.e., path homology with dimension equal to the maximum path length of a graph. Numerical exper- iments show that the algorithm has a significant advantage over the general algorithm in computation time as the depth of stratified digraph increases.

cs.CG↗

arXiv2022

Auto-Encoding Goodness of Fit

We develop a new type of generative autoencoder called the Goodness-of-Fit Autoencoder (GoFAE), which incorporates GoF tests at two levels. At the minibatch level, it uses GoF test statistics as regularization objectives. At a more global level, it selects a regularization coefficient based on higher criticism, i.e., a test on the uniformity of the local GoF p-values. We justify the use of GoF tests by providing a relaxed $L_2$-Wasserstein bound on the distance between the latent distribution and a distribution class. We prove that optimization based on these tests can be done with stochastic gradient descent on a compact Riemannian manifold. Empirically, we show that our higher criticism parameter selection procedure balances reconstruction and generation using mutual information and uniformity of p-values respectively. Finally, we show that GoFAE achieves comparable FID scores and mean squared errors with competing deep generative models while retaining statistical indistinguishability from Gaussian in the latent space based on a variety of hypothesis tests.

cs.LG↗

arXiv2020

Spherical harmonic analysis for multivariate stable distributions

Series representations consisting of spherical harmonics are obtained for characteristic exponents and probability density functions of multivariate stable distributions under various conditions. A esult potentially applicable in a practical setting is that for any distribution with stability index not equal to 1 and with a polynomial spectral spherical density, the series representation converges absolutely with all terms being calculable in closed form. Asymptotic expansions consisting of spherical harmonics are also considered for probability density functions.

math.PR↗

arXiv2018

Law of two-sided exit by a spectrally positive strictly stable process

For a spectrally positive strictly stable process with index in (1,2), the paper obtains i) the density of the time when the process makes first exit from an interval by hitting the interval's lower end point before jumping over its upper end point, and ii) the joint distribution of the time, the undershoot, and the jump of the process when it makes first exit the other way around. For i), the density of the time of first exit is expressed as an infinite sum of functions, each the product of a polynomial and an exponential function, with all coefficients determined by the roots of a Mittag-Leffler function. For ii), conditional on the undershoot, the time and the jump of first exit are independent, and the marginal conditional densities of the time has similar features as i).

math.PR↗

arXiv2018

Law of the first passage triple of a spectrally positive strictly stable process

For a spectrally positive and strictly stable process with index in (1,2), a series representation is obtained for the joint distribution of the "first passage triple" that consists of the time of first passage and the undershoot and the overshoot at first passage. The result leads to several corollaries, including 1) the joint law of the first passage triple and the pre-passage running supremum, and 2) at a fixed time point, the joint law of the process' value, running supremum, and the time of the running supremum. The representation can be decomposed as a sum of strictly positive functions that allows exact sampling of the first passage triple.

math.PR↗

arXiv2016

Exact sampling of first passage event of certain symmetric Levy processes with unbounded variation

We show that exact sampling of the first passage event can be done for a Levy process with unbounded variation, if the process can be embedded in a subordinated standard Brownian motion. By sampling a series of first exit events of the Brownian motion and first passage events of the subordinator, the first passage event of interest can be obtained. The sampling of the first exit time and pre-exit location of the Brownian motion may be of independent interest.

math.PR↗

arXiv2015

Strong renewal theorems with infinite mean beyond local large deviations

Let $F$ be a distribution function on the line in the domain of attraction of a stable law with exponent $\alpha\in(0,1/2]$. We establish the strong renewal theorem for a random walk $S_1,S_2,\ldots$ with step distribution $F$, by extending the large deviations approach in Doney [Probab. Theory Related Fileds 107 (1997) 451-465]. This is done by introducing conditions on $F$ that in general rule out local large deviations bounds of the type $\mathbb{P}\{S_n\in(x,x+h]\}=O(n)\overline{F}(x)/x$, hence are significantly weaker than the boundedness condition in Doney (1997). We also give applications of the results on ladder height processes and infinitely divisible distributions.

math.PR↗

arXiv2015

Random reversible Markov matrices with tunable extremal eigenvalues

Random sampling of large Markov matrices with a tunable spectral gap, a nonuniform stationary distribution, and a nondegenerate limiting empirical spectral distribution (ESD) is useful. Fix $c>0$ and $p>0$. Let $A_n$ be the adjacency matrix of a random graph following $\mathrm{G}(n, p/n)$, known as the Erd\H{o}s-R\'enyi distribution. Add $c/n$ to each entry of $A_n$ and then normalize its rows. It is shown that the resulting Markov matrix has the desired properties. Its ESD weakly converges in probability to a symmetric nondegenerate distribution, and its extremal eigenvalues, other than 1, fall in $[-1/\sqrt{1+c/k},-b]\cup [b,1/\sqrt{1+c/k}]$ for any $0< b < 1/\sqrt{1+c}$, where $k = \lfloor p \rfloor + 1$. Thus, for $p\in (0,1)$, the spectral gap tends to $1-1/\sqrt{1+c}$.

math.PR↗

arXiv2014

False Discovery Variance Reduction in Large Scale Simultaneous Hypothesis Tests

Statistical dependence between hypotheses poses a significant challenge to the stability of large scale multiple hypotheses testing. Ignoring it often results in an unacceptably large spread in the false positive proportion even though the average value is acceptable [21, 39, 40, 49]. However, the statistical dependence structure of data is often unknown. Using a generic signalprocessing model, Bayesian multiple testing, and simulations, we demonstrate that the variance of the false positive proportion can be substantially reduced even under unknown short range dependence. We do this by modeling the data generating process as a stationary ergodic binary signal process embedded in noisy observations. We derive conditional probabilities needed for the Bayesian multiple testing by incorporating nearby observations into a second order Taylor series approximation. Simulations under general conditions are carried out to assess the validity and the variance reduction of the approach. Along the way, we address the problem of sampling a random Markov matrix with specified stationary distribution and lower bounds on the top absolute eigenvalues, which is of interest in its own right.

stat.ME↗

arXiv2014

On Multivariate Strong Renewal Theorem

This paper takes the so-called probabilistic approach to the Strong Renewal Theorem (SRT) for multivariate distributions in the domain of attraction of a stable law. A version of the SRT is obtained that allows any kind of lattice-nonlattice composition of a distribution. A general bound is derived to control the so-called "small-$n$ contribution", which arises from random walk paths that have a relatively small number of steps but make large cumulative moves. The asymptotic negligibility of the small-$n$ contribution is essential to the SRT. Applications of the SRT are given, including some that provide a unified treatment to known results but with substantially weaker assumptions.

math.PR↗

arXiv2013

Integral criteria for Strong Renewal Theorems with infinite mean

Let $F$ be a probability measure on $\mathbb{R}$ in the domain of attraction of a stable law with exponent $\alpha\in (0, 1)$. We establish integral criteria on $F$ that significantly expand the probabilistic approach to Strong Renewal Theorems (SRTs). The criterion for $\alpha \in (0,1/2]$ is much weaker than currently available ones and in some cases provides sufficient and necessary conditions for the SRT. The criterion for $\alpha \in (1/2, 1)$ establishes the SRT in full generality and in a unified way, barring the Limit Local Theorems employed. As an application, for infinitely divisible $F$, an integral criterion on its L\'evy measure is established for the SRT. As another application, for $F$ in the domain of attraction of a stable law without centering, an integral criterion on $F$ is established for the SRT for the ladder height process of a random walk with step distribution $F$.

math.PR↗

arXiv2013

Nonnormal small jump approximation of infinitely divisible distributions

We consider a type of nonnormal approximation of infinitely divisible distributions that incorporates compound Poisson, Gamma, and normal distributions. The approximation relies on achieving higher orders of cumulant matching, to obtain higher rates of approximation error decay. The parameters of the approximation are easy to fix. The computational complexity of random sampling of the approximating distribution in many cases is of the same order as normal approximation. Error bounds in terms of total variance distance are derived. Both the univariate and the multivariate cases of the approximation are considered.

math.PR↗

arXiv2012

On exact sampling of the first passage event of Levy process with infinite Levy measure and bounded variation

We present an exact sampling method for the first passage event of a Levy process. The idea is to embed the process into another one whose first passage event can be sampled exactly, and then recover the part belonging to the former from the latter. The method is based on several distributional properties that appear to be new. We obtain general procedures to sample the first passage event of a subordinator across a regular non-increasing boundary, and that of a process with infinite Levy measure, bounded variation, and suitable drift across a constant level or interval. We give examples of application to a rather wide variety of Levy measures.

math.PR↗

arXiv2010

Stochastic Lipschitz continuity for high dimensional Lasso with multiple linear covariate structures or hidden linear covariates

Two extensions of generalized linear models are considered. In the first one, response variables depend on multiple linear combinations of covariates. In the second one, only response variables are observed while the linear covariates are missing. We derive stochastic Lipschitz continuity results for the loss functions involved in the regression problems and apply them to get bounds on estimation error for Lasso. Multivariate comparison results on Rademacher complexity are obtained as tools to establish the stochastic Lipschitz continuity results.

math.ST↗