arXiv subjects

Tingzhou Yu

Publications and source records attributed to Tingzhou Yu.

5 recordsLinked to original sources

arXiv2026

Stochastic Rounding Increases Small Singular Values

Over the past half-dozen years, stochastic rounding (SR) has regained significant attention as a quantization scheme for low-precision floating-point arithmetic, with applications spanning numerical analysis and modern machine learning systems. Recent work has shown that SR acts as an implicit regularizer by increasing the smallest singular value of extremely tall-and-thin (or, symmetrically, short-and-fat) matrices. In this work, we substantially sharpen and extend this understanding in two directions. First, we show that the regularization effect of SR is not restricted to extreme aspect ratio regimes: it persists for matrices with constant aspect ratio. Second, we demonstrate that SR does not merely regularize the smallest singular value, but instead lifts entire clusters of singular values at the tail of the spectrum. Together, these results provide a more general characterization of stochastic rounding as a spectral regularizer, revealing that its effects extend beyond extremal aspect ratios and act on a broader portion of the singular value spectrum.

math.NA↗

arXiv2026

An upper bound on the smallest singular value of dense random combinatorial matrices

Let $M$ be an $n\times n$ random matrix with entries in $\{0, 1\}$, where each row is independently and uniformly sampled from the set of all vectors in $\{0, 1\}^n$ containing exactly $d$ ones, with $d=pn$ for some fixed constant $p\in (0,1/2]$. A recent result of Tran states that the smallest singular value $s_n(M)$ is bounded below by $c_p n^{-1/2}$ with high probability. In this note, we establish a complementary upper bound for $s_n(M)$, proving that \[ \forall \varepsilon >0 \qquad \mathbb{P}\left(s_n(M)\le \frac{\sqrt{d}}{\varepsilon^2 n}\right)\ge 1-C_p\left(\varepsilon+\frac{1}{\sqrt{d}}\right), \]where $C_p$ is a positive constant depending only on $p$. This result confirms that the least singular value $s_n(M)$ of dense random combinatorial matrices is typically of the order $n^{-1/2}$.

math.PR↗

arXiv2026

Least singular values of shifted sparse random combinatorial matrices

Let $M_n$ be an $n\times n$ random matrix with entries in $\{0, 1\}$, where each row is independently and uniformly sampled from the set of all vectors in $\{0, 1\}^n$ containing exactly $d$ ones. we establish quantitative lower bounds on the smallest singular value of the shifted matrices $M_n-z \mathbf{I}_n$ whenever $|z| \leq \sqrt{d}\, \log\log d$ and $ C \log n \leq d \leq n/2$ for some absolute positive constant $C$. As an application, we show that the empirical spectral distribution of the appropriately rescaled matrix $M_n$ converges in probability to the circular law provided that $\log^{2+\varepsilon} n \leq d=o(n)$ for some fixed $\varepsilon \in (0,1)$.

math.PR↗

arXiv2024

Analyzing dynamics and average case complexity in the spherical Sherrington-Kirkpatrick model: a focus on extreme eigenvectors

We explore Langevin dynamics in the spherical Sherrington-Kirkpatrick model, delving into the asymptotic energy limit. Our approach involves integro-differential equations, incorporating the Crisanti-Horner-Sommers-Cugliandolo-Kurchan equation from spin glass literature, to analyze the system's size and its temperature-dependent phase transition. Additionally, we conduct an average case complexity analysis, establishing hitting time bounds for the bottom eigenvector of a Wigner matrix. Our investigation also includes the power iteration algorithm, examining its average case complexity in identifying the top eigenvector overlap, with comprehensive complexity bounds.

math.PR↗

arXiv2021

Quantitative Russo-Seymour-Welsh for random walk on random graphs and decorrelation of UST

We prove a quantitative Russo-Seymour-Welsh (RSW) type result for random walks on two natural examples of random planar graphs: the supercritical percolation cluster in the square lattice and the Poisson Voronoi triangulation in the plane. More precisely, we prove that the probability that a simple random walk crosses a rectangle in the hard direction with uniformly positive probability is stretched exponentially likely in the size of the rectangle. As an application we prove a near optimal decorrelation result for uniform spanning trees for such graphs. This is the key missing step in this setup while applying of the proof stretegy of a previous article on universality of dimers ("Dimers and imaginary geometry." Ann. Probab. 48 (1) 1 - 52) where random walk RSW was assumed to hold with probability 1. Applications to almost sure Gaussian free field scaling limit for dimers on Temperleyan type modification on such graphs are also discussed.

math.PR↗