arXiv subjects

Yazhen Wang

Publications and source records attributed to Yazhen Wang.

At least 19 recordsLinked to original sources

arXiv2026

Statistical Analysis of Quantum Annealing

Quantum computers use quantum resources to carry out computational tasks and may outperform classical computers in solving certain computational problems. Special-purpose quantum computers such as quantum annealers employ quantum adiabatic theorem to solve combinatorial optimization problems. In this paper, we compare classical annealings such as simulated annealing and quantum annealings that are done by the D-Wave machines both theoretically and numerically. We show that if the classical and quantum annealing are characterized by equivalent Ising models, then solving an optimization problem, i.e., finding the minimal energy of each Ising model, by the two annealing procedures, are mathematically identical. For quantum annealing, we also derive the probability lower-bound on successfully solving an optimization problem by measuring the system at the end of the annealing procedure. Moreover, we present the Markov chain Monte Carlo (MCMC) method to realize quantum annealing by classical computers and investigate its statistical properties. In the numerical section, we discuss the discrepancies between the MCMC based annealing approaches and the quantum annealing approach in solving optimization problems.

stat.OT↗

arXiv2026

SADA: Safe and Adaptive Aggregation of Multiple Black-Box Predictions in Semi-Supervised Learning

Semi-supervised learning (SSL) arises in practice when labeled data are scarce or expensive to obtain, while large quantities of unlabeled data are readily available. With the growing adoption of machine learning techniques, it has become increasingly feasible to generate multiple predicted labels using a variety of models and algorithms, including deep learning, large language models, and generative AI. In this paper, we propose a novel approach that safely and adaptively aggregates multiple black-box predictions of uncertain quality for both inference and prediction tasks. Our method provides two key guarantees: (i) it never performs worse than using the labeled data alone, regardless of the quality of the predictions; and (ii) if any one of the predictions (without knowing which one) perfectly fits the ground truth, the algorithm adaptively exploits this to achieve either a faster convergence rate or the semiparametric efficiency bound. We demonstrate the effectiveness of the proposed algorithm through small-scale simulations and two real-data analyses with distinct scientific goals. A user-friendly R package, sada, is provided to facilitate practical implementation.

stat.ML↗

arXiv2026

Harnessing Bayesian Statistics to Accelerate Iterative Quantum Amplitude Estimation

We establish a unified statistical framework that underscores the crucial role statistical inference plays in Quantum Amplitude Estimation (QAE), a task essential to fields ranging from chemistry to finance and machine learning. We use this framework to harness Bayesian statistics for improved measurement efficiency with rigorous interval estimates at all iterations of Iterative Quantum Amplitude Estimation. We demonstrate the resulting method, Bayesian Iterative Quantum Amplitude Estimation (BIQAE), accurately and efficiently estimates both quantum amplitudes and molecular ground-state energies to high accuracy, and show in analytic and numerical sample complexity analyses that BIQAE outperforms all other QAE approaches considered. Both rigorous mathematical proofs and numerical simulations conclusively indicate Bayesian statistics is the source of this advantage, a finding that invites further inquiry into the power of statistics to expedite the search for quantum utility.

quant-ph↗

arXiv2026

Learning Relationship between Quantum Walks and Underdamped Langevin Dynamics

Fast computational algorithms are in constant demand, and their development has been driven by advances such as quantum speedup and classical acceleration. This paper intends to study search algorithms based on quantum walks in quantum computation and sampling algorithms based on Langevin dynamics in classical computation. On the quantum side, quantum walk-based search algorithms can achieve quadratic speedups over their classical counterparts. In classical computation, a substantial body of work has focused on gradient acceleration, with gradient-adjusted algorithms derived from underdamped Langevin dynamics providing quadratic acceleration over conventional Langevin algorithms. Since both search and sampling algorithms are designed to address learning tasks, we study learning relationship between coined quantum walks and underdamped Langevin dynamics. Specifically, we show that, in terms of the Le Cam deficiency distance, a quantum walk with randomization is asymptotically equivalent to underdamped Langevin dynamics, whereas the quantum walk without randomization is not asymptotically equivalent due to its high-frequency oscillatory behavior. We further discuss the implications of these equivalence and nonequivalence results for the computational and inferential properties of the associated algorithms in machine learning tasks. Our findings offer new insight into the relationship between quantum walks and underdamped Langevin dynamics, as well as the intrinsic mechanisms underlying quantum speedup and classical gradient acceleration.

quant-ph↗

arXiv2025

Factor and Idiosyncratic VAR Volatility Matrix Models for Heavy-Tailed High-Frequency Financial Observations

This paper introduces a novel process for both factor and idiosyncratic volatility matrices whose eigenvalues follow the vector auto-regressive (VAR) model. We call it the factor and idiosyncratic VAR (FIVAR) model. The FIVAR model accounts for the dynamics of the factor and idiosyncratic volatilities and includes many parameters. In addition, many empirical studies have shown that high-frequency stock returns and volatilities often exhibit heavy tails. To handle these two problems simultaneously, we propose a penalized optimization procedure with a truncation scheme for parameter estimation. We apply the proposed parameter estimation procedure to predicting large volatility matrices and establish its asymptotic properties.

stat.ME↗

arXiv2025

Robust Reinforcement Learning under Diffusion Models for Data with Jumps

Reinforcement Learning (RL) has proven effective in solving complex decision-making tasks across various domains, but challenges remain in continuous-time settings, particularly when state dynamics are governed by stochastic differential equations (SDEs) with jump components. In this paper, we address this challenge by introducing the Mean-Square Bipower Variation Error (MSBVE) algorithm, which enhances robustness and convergence in scenarios involving significant stochastic noise and jumps. We first revisit the Mean-Square TD Error (MSTDE) algorithm, commonly used in continuous-time RL, and highlight its limitations in handling jumps in state dynamics. The proposed MSBVE algorithm minimizes the mean-square quadratic variation error, offering improved performance over MSTDE in environments characterized by SDEs with jumps. Simulations and formal proofs demonstrate that the MSBVE algorithm reliably estimates the value function in complex settings, surpassing MSTDE's performance when faced with jump processes. These findings underscore the importance of alternative error metrics to improve the resilience and effectiveness of RL algorithms in continuous-time frameworks.

cs.LG↗

arXiv2025

Computational and Statistical Asymptotic Analysis of the JKO Scheme for Iterative Algorithms to update distributions

The seminal paper of Jordan, Kinderlehrer, and Otto introduced what is now widely known as the JKO scheme, an iterative algorithmic framework for computing distributions. This scheme can be interpreted as a Wasserstein gradient flow and has been successfully applied in machine learning contexts, such as deriving policy solutions in reinforcement learning. In this paper, we extend the JKO scheme to accommodate models with unknown parameters. Specifically, we develop statistical methods to estimate these parameters and adapt the JKO scheme to incorporate the estimated values. To analyze the adopted statistical JKO scheme, we establish an asymptotic theory via stochastic partial differential equations that describes its limiting dynamic behavior. Our framework allows both the sample size used in parameter estimation and the number of algorithmic iterations to go to infinity. This study offers a unified framework for joint computational and statistical asymptotic analysis of the statistical JKO scheme. On the computational side, we examine the scheme's dynamic behavior as the number of iterations increases, while on the statistical side, we investigate the large-sample behavior of the resulting distributions computed through the scheme. We conduct numerical simulations to evaluate the finite-sample performance of the proposed methods and validate the developed asymptotic theory.

stat.ML↗

arXiv2024

Robust Realized Integrated Beta Estimator with Application to Dynamic Analysis of Integrated Beta

In this paper, we develop a robust non-parametric realized integrated beta estimator using high-frequency financial data contaminated by microstructure noises, which is robust to the stylized features, such as the time-varying beta and the dependence structure of microstructure noises. With this robust realized integrated beta estimator, we investigate dynamic structures of integrated betas and find an auto-regressive--moving-average (ARMA) structure. To model this dynamic structure, we utilize the ARMA model for daily integrated market betas. We call this the dynamic realized beta (DR Beta). We further introduce a high-frequency data generating process by filling the gap between the high-frequency-based non-parametric estimator and low-frequency dynamic structure. Then, we propose a quasi-likelihood procedure for estimating the model parameters with the robust realized integrated beta estimator as the proxy. We also establish asymptotic theorems for the proposed estimator and conduct a simulation study to check the performance of finite samples of the estimator. The empirical study with the S&P 500 index and the top 50 large trading volume stocks from the S&P 500 illustrates that the proposed DR Beta model with the robust realized beta estimator effectively accounts for dynamics in the market beta of individual stocks and better predicts future market betas.

stat.ME↗

arXiv2023

Factor Overnight GARCH-Itô Models

This paper introduces a unified factor overnight GARCH-Itô model for large volatility matrix estimation and prediction. To account for whole-day market dynamics, the proposed model has two different instantaneous factor volatility processes for the open-to-close and close-to-open periods, while each embeds the discrete-time multivariate GARCH model structure. To estimate latent factor volatility, we assume the low rank plus sparse structure and employ nonparametric estimation procedures. Then, based on the connection between the discrete-time model structure and the continuous-time diffusion process, we propose a weighted least squares estimation procedure with the non-parametric factor volatility estimator and establish its asymptotic theorems.

stat.ME↗

arXiv2022

Overnight GARCH-Itô Volatility Models

Various parametric volatility models for financial data have been developed to incorporate high-frequency realized volatilities and better capture market dynamics. However, because high-frequency trading data are not available during the close-to-open period, the volatility models often ignore volatility information over the close-to-open period and thus may suffer from loss of important information relevant to market dynamics. In this paper, to account for whole-day market dynamics, we propose an overnight volatility model based on Itô diffusions to accommodate two different instantaneous volatility processes for the open-to-close and close-to-open periods. We develop a weighted least squares method to estimate model parameters for two different periods and investigate its asymptotic properties. We conduct a simulation study to check the finite sample performance of the proposed model and method. Finally, we apply the proposed approaches to real trading data.

q-fin.ST↗

arXiv2022

Dynamic Realized Beta Models Using Robust Realized Integrated Beta Estimator

This paper introduces a unified parametric modeling approach for time-varying market betas that can accommodate continuous-time diffusion and discrete-time series models based on a continuous-time series regression model to better capture the dynamic evolution of market betas. We call this the dynamic realized beta (DR Beta). We first develop a non-parametric realized integrated beta estimator using high-frequency financial data contaminated by microstructure noises, which is robust to the stylized features, such as the time-varying beta and the dependence structure of microstructure noises, and construct the estimator's asymptotic properties. Then, with the robust realized integrated beta estimator, we propose a quasi-likelihood procedure for estimating the model parameters based on the combined high-frequency data and low frequency dynamic structure. We also establish asymptotic theorems for the proposed estimator and conduct a simulation study to check the performance of finite samples of the estimator. The empirical study with the S&P 500 index and the top 50 large trading volume stocks from the S&P 500 illustrates that the proposed DR Beta model effectively accounts for dynamics in the market beta of individual stocks and better predicts future market betas.

stat.ME↗

arXiv2022

Optimal High-order Tensor SVD via Tensor-Train Orthogonal Iteration

This paper studies a general framework for high-order tensor SVD. We propose a new computationally efficient algorithm, tensor-train orthogonal iteration (TTOI), that aims to estimate the low tensor-train rank structure from the noisy high-order tensor observation. The proposed TTOI consists of initialization via TT-SVD (Oseledets, 2011) and new iterative backward/forward updates. We develop the general upper bound on estimation error for TTOI with the support of several new representation lemmas on tensor matricizations. By developing a matching information-theoretic lower bound, we also prove that TTOI achieves the minimax optimality under the spiked tensor model. The merits of the proposed TTOI are illustrated through applications to estimation and dimension reduction of high-order Markov processes, numerical studies, and a real data example on New York City taxi travel records. The software of the proposed algorithm is available online$^6$.

math.ST↗

arXiv2021

Conditional Quantile Analysis for Realized GARCH Models

This paper introduces a novel quantile approach to harness the high-frequency information and improve the daily conditional quantile estimation. Specifically, we model the conditional standard deviation as a realized GARCH model and employ conditional standard deviation, realized volatility, realized quantile, and absolute overnight return as innovations in the proposed dynamic quantile models. We devise a two-step estimation procedure to estimate the conditional quantile parameters. The first step applies a quasi-maximum likelihood estimation procedure, with the realized volatility as a proxy for the volatility proxy, to estimate the conditional standard deviation parameters. The second step utilizes a quantile regression estimation procedure with the estimated conditional standard deviation in the first step. Asymptotic theory is established for the proposed estimation methods, and a simulation study is conducted to check their finite-sample performance. Finally, we apply the proposed methodology to calculate the value at risk (VaR) of 20 individual assets and compare its performance with existing competitors.

stat.ME↗

arXiv2020

SGD Distributional Dynamics of Three Layer Neural Networks

With the rise of big data analytics, multi-layer neural networks have surfaced as one of the most powerful machine learning methods. However, their theoretical mathematical properties are still not fully understood. Training a neural network requires optimizing a non-convex objective function, typically done using stochastic gradient descent (SGD). In this paper, we seek to extend the mean field results of Mei et al. (2018) from two-layer neural networks with one hidden layer to three-layer neural networks with two hidden layers. We will show that the SGD dynamics is captured by a set of non-linear partial differential equations, and prove that the distributions of weights in the two hidden layers are independent. We will also detail exploratory work done based on simulation and real-world data.

cs.LG↗

arXiv2020

How Many Factors Influence Minima in SGD?

Stochastic gradient descent (SGD) is often applied to train Deep Neural Networks (DNNs), and research efforts have been devoted to investigate the convergent dynamics of SGD and minima found by SGD. The influencing factors identified in the literature include learning rate, batch size, Hessian, and gradient covariance, and stochastic differential equations are used to model SGD and establish the relationships among these factors for characterizing minima found by SGD. It has been found that the ratio of batch size to learning rate is a main factor in highlighting the underlying SGD dynamics; however, the influence of other important factors such as the Hessian and gradient covariance is not entirely agreed upon. This paper describes the factors and relationships in the recent literature and presents numerical findings on the relationships. In particular, it confirms the four-factor and general relationship results obtained in Wang (2019), while the three-factor and associated relationship results found in Jastrzȩbski et al. (2018) may not hold beyond the considered special case.

cs.LG↗

arXiv2020

Unified Discrete-Time Factor Stochastic Volatility and Continuous-Time Ito Models for Combining Inference Based on Low-Frequency and High-Frequency

This paper introduces unified models for high-dimensional factor-based Ito process, which can accommodate both continuous-time Ito diffusion and discrete-time stochastic volatility (SV) models by embedding the discrete SV model in the continuous instantaneous factor volatility process. We call it the SV-Ito model. Based on the series of daily integrated factor volatility matrix estimators, we propose quasi-maximum likelihood and least squares estimation methods. Their asymptotic properties are established. We apply the proposed method to predict future vast volatility matrix whose asymptotic behaviors are studied. A simulation study is conducted to check the finite sample performance of the proposed estimation and prediction method. An empirical analysis is carried out to demonstrate the advantage of the SV-Ito model in volatility prediction and portfolio allocation problems.

stat.ME↗

arXiv2020

Volatility Analysis with Realized GARCH-Ito Models

This paper introduces a unified approach for modeling high-frequency financial data that can accommodate both the continuous-time jump-diffusion and discrete-time realized GARCH model by embedding the discrete realized GARCH structure in the continuous instantaneous volatility process. The key feature of the proposed model is that the corresponding conditional daily integrated volatility adopts an autoregressive structure where both integrated volatility and jump variation serve as innovations. We name it as the realized GARCH-Ito model. Given the autoregressive structure in the conditional daily integrated volatility, we propose a quasi-likelihood function for parameter estimation and establish its asymptotic properties. To improve the parameter estimation, we propose a joint quasi-likelihood function that is built on the marriage of daily integrated volatility estimated by high-frequency data and nonparametric volatility estimator obtained from option data. We conduct a simulation study to check the finite sample performance of the proposed methodologies and an empirical study with the S&P500 stock index and option data.

stat.ME↗

arXiv2019

Asymptotic Analysis via Stochastic Differential Equations of Gradient Descent Algorithms in Statistical and Computational Paradigms

This paper investigates asymptotic behaviors of gradient descent algorithms (particularly accelerated gradient descent and stochastic gradient descent) in the context of stochastic optimization arising in statistics and machine learning where objective functions are estimated from available data. We show that these algorithms can be computationally modeled by continuous-time ordinary or stochastic differential equations. We establish gradient flow central limit theorems to describe the limiting dynamic behaviors of these computational algorithms and the large-sample performances of the related statistical procedures, as the number of algorithm iterations and data size both go to infinity, where the gradient flow central limit theorems are governed by some linear ordinary or stochastic differential equations like time-dependent Ornstein-Uhlenbeck processes. We illustrate that our study can provide a novel unified framework for a joint computational and statistical asymptotic analysis, where the computational asymptotic analysis studies dynamic behaviors of these algorithms with the time (or the number of iterations in the algorithms), the statistical asymptotic analysis investigates large sample behaviors of the statistical procedures (like estimators and classifiers) that the algorithms are applied to compute, and in fact the statistical procedures are equal to the limits of the random sequences generated from these iterative algorithms as the number of iterations goes to infinity. The joint analysis results based on the obtained gradient flow central limit theorems can identify four factors - learning rate, batch size, gradient covariance, and Hessian - to derive new theory regarding the local minima found by stochastic gradient descent for solving non-convex optimization problems.

stat.ML↗