High Probability Streaming Lower Bounds for $F_2$ Estimation
本文解决了流模型中$F_2$估计问题,通过引入噪声鲁棒通信原语并设计两种算法,在不同假设下优化了空间复杂度与依赖性。
本文解决了流模型中$F_2$估计问题,通过引入噪声鲁棒通信原语并设计两种算法,在不同假设下优化了空间复杂度与依赖性。
该研究通过在函数空间中直接开发统计力学描述来探讨深度神经网络的学习动态问题,使用了学习算子和统计算子等方法来解析误差动态和参数空间的状态密度。
This work refutes Steurer’s conjecture that any family of unit vectors with low pairwise correlations must contain a constant-separated subset of linear size. By constructing a counterexample based on sparse high-dimensional expanders, the authors demonstrate that even when the average correlation among vectors is polynomially small, no linear-sized approximately separated subset may exist. The proof hinges on three key technical contributions: the design of high-dimensional expanders, a refined analysis of vector correlations, and an estimate of the $L_2$ mixing time for reweighted random walks. As a byproduct, this construction yields the first example of a vertex expander on which every reweighted simple random walk exhibits an $L_2$ mixing time of at least $\log^{5/4 - o(1)} n$, thereby revealing a fundamental limitation in the relationship between correlation and separability.
This work establishes the optimal approximation ratio for computing the permanent of Hermitian positive semidefinite matrices in deterministic polynomial time. By formulating a concave optimization problem based on the row vectors of a matrix factorization, analyzing the Wick integral formula via entropy methods, and leveraging complex matrix decomposition techniques, the authors derive tight upper and lower bounds on the permanent. Their main contribution is the first complete characterization of the optimal approximation ratio at exponential precision, proving that $\operatorname{per}(A)$ satisfies $e^{-\gamma n} \widehat P(A) \le \operatorname{per}(A) \le \widehat P(A)$. This yields an approximation ratio of $e^{(\gamma+o(1))n}$, which exactly matches the known computational hardness lower bound, thereby closing a longstanding gap in the theoretical understanding of this problem.
This paper addresses the statistical inference challenge for Markov chain-driven reinforcement learning (e.g., temporal difference learning) under data dependence. It establishes, for the first time, concentration inequalities and Berry–Esseen-type bounds on distributional convergence rates for high-dimensional vector- and matrix-valued functions of Markov chains. Methodologically, it integrates stochastic process theory, spectral analysis, and high-dimensional probability tools to derive high-probability consistency guarantees that match the asymptotic variance, achieving a Gaussian approximation rate of $O(T^{-1/4}log T)$ in convex distance. The key contributions are: (1) the first non-asymptotic, high-probability convergence guarantee for TD estimators under non-i.i.d. data—sharper than prior results; and (2) a rigorous theoretical foundation for uncertainty quantification and statistical inference in reinforcement learning.
本文解决了流模型中$F_2$估计问题,通过引入噪声鲁棒通信原语并设计两种算法,在不同假设下优化了空间复杂度与依赖性。
该研究通过在函数空间中直接开发统计力学描述来探讨深度神经网络的学习动态问题,使用了学习算子和统计算子等方法来解析误差动态和参数空间的状态密度。
This work refutes Steurer’s conjecture that any family of unit vectors with low pairwise correlations must contain a constant-separated subset of linear size. By constructing a counterexample based on sparse high-dimensional expanders, the authors demonstrate that even when the average correlation among vectors is polynomially small, no linear-sized approximately separated subset may exist. The proof hinges on three key technical contributions: the design of high-dimensional expanders, a refined analysis of vector correlations, and an estimate of the $L_2$ mixing time for reweighted random walks. As a byproduct, this construction yields the first example of a vertex expander on which every reweighted simple random walk exhibits an $L_2$ mixing time of at least $\log^{5/4 - o(1)} n$, thereby revealing a fundamental limitation in the relationship between correlation and separability.
This work establishes the optimal approximation ratio for computing the permanent of Hermitian positive semidefinite matrices in deterministic polynomial time. By formulating a concave optimization problem based on the row vectors of a matrix factorization, analyzing the Wick integral formula via entropy methods, and leveraging complex matrix decomposition techniques, the authors derive tight upper and lower bounds on the permanent. Their main contribution is the first complete characterization of the optimal approximation ratio at exponential precision, proving that $\operatorname{per}(A)$ satisfies $e^{-\gamma n} \widehat P(A) \le \operatorname{per}(A) \le \widehat P(A)$. This yields an approximation ratio of $e^{(\gamma+o(1))n}$, which exactly matches the known computational hardness lower bound, thereby closing a longstanding gap in the theoretical understanding of this problem.
This paper addresses the statistical inference challenge for Markov chain-driven reinforcement learning (e.g., temporal difference learning) under data dependence. It establishes, for the first time, concentration inequalities and Berry–Esseen-type bounds on distributional convergence rates for high-dimensional vector- and matrix-valued functions of Markov chains. Methodologically, it integrates stochastic process theory, spectral analysis, and high-dimensional probability tools to derive high-probability consistency guarantees that match the asymptotic variance, achieving a Gaussian approximation rate of $O(T^{-1/4}log T)$ in convex distance. The key contributions are: (1) the first non-asymptotic, high-probability convergence guarantee for TD estimators under non-i.i.d. data—sharper than prior results; and (2) a rigorous theoretical foundation for uncertainty quantification and statistical inference in reinforcement learning.