Survey of Novel Deep Learning Architectures for Denoising Gravitational-wave Signals
研究通过比较五种深度学习架构解决引力波信号去噪问题，提出多尺度频率感知架构，在全参数空间实现高效去噪。
0 citationsRead paper
Selected work
Representative Papers
Authentication over Arbitrarily Varying Channels with Causal Adversaries
本文研究了在具有因果对手的任意变化信道上进行身份验证的问题，通过构建正速率码和使用martingale集中论点来解决该问题。
0 citationsRead paper
Algorithms for Robbins' Problem using Markov Decision Processes
本文针对Robbins问题，通过建模为马尔可夫决策过程并提出有限状态抽象方法，以近似解决在已知全部信息下最小化选择候选人期望排名的问题。
0 citationsRead paper
Simultaneous Coverage and Efficiency Guarantee in Online Conformal Prediction
This work addresses a critical limitation of existing online conformal prediction methods, which under distribution shift only control signed coverage error and thus risk persistent one-sided miscoverage while lacking effective constraints on prediction set size. The paper proposes a unified online learning framework that simultaneously guarantees absolute coverage validity and prediction efficiency across three settings: adversarial, stochastic, and covariate-dependent shifts. Its core contributions include the first joint control of non-canceling coverage bias and efficiency; the design of a sliding-window quantile tracker combined with a partitioned ACI algorithm; and matching minimax lower bounds establishing rate optimality. The method provides distribution-free, convexity-free theoretical guarantees, achieves optimal convergence rates in the stochastic setting, and ensures simultaneous coverage and efficiency relative to dynamic oracle thresholds under covariate-dependent shifts.
0 citationsRead paper
Deterministic Algorithms for Low Individual Degree Factors of Sparse Polynomials
This work addresses the problem of efficiently and deterministically enumerating all factors of bounded individual degree in sparse polynomials. For an $n$-variate, $s$-sparse polynomial with individual degrees at most $d$, the paper presents the first deterministic quasi-polynomial time algorithm that works over arbitrary fields and accommodates unbounded total degree. The approach combines interpolation, divisibility testing, and algebraic complexity techniques to construct constant-depth arithmetic circuits capturing all such factors, achieving a running time of $\mathrm{poly}(n, s^d)$. For general sparse polynomials, the algorithm runs in time $\mathrm{poly}(D^{d \log s}, s^{d^2 \log n})$, where $D$ bounds the total degree. These results generalize prior work and yield new upper bounds on the number of low individual-degree factors.
0 citationsRead paper
Recent publications
Latest Papers
Survey of Novel Deep Learning Architectures for Denoising Gravitational-wave Signals
研究通过比较五种深度学习架构解决引力波信号去噪问题，提出多尺度频率感知架构，在全参数空间实现高效去噪。
0 citationsRead paper
Authentication over Arbitrarily Varying Channels with Causal Adversaries
本文研究了在具有因果对手的任意变化信道上进行身份验证的问题，通过构建正速率码和使用martingale集中论点来解决该问题。
0 citationsRead paper
Algorithms for Robbins' Problem using Markov Decision Processes
本文针对Robbins问题，通过建模为马尔可夫决策过程并提出有限状态抽象方法，以近似解决在已知全部信息下最小化选择候选人期望排名的问题。
0 citationsRead paper
Simultaneous Coverage and Efficiency Guarantee in Online Conformal Prediction
This work addresses a critical limitation of existing online conformal prediction methods, which under distribution shift only control signed coverage error and thus risk persistent one-sided miscoverage while lacking effective constraints on prediction set size. The paper proposes a unified online learning framework that simultaneously guarantees absolute coverage validity and prediction efficiency across three settings: adversarial, stochastic, and covariate-dependent shifts. Its core contributions include the first joint control of non-canceling coverage bias and efficiency; the design of a sliding-window quantile tracker combined with a partitioned ACI algorithm; and matching minimax lower bounds establishing rate optimality. The method provides distribution-free, convexity-free theoretical guarantees, achieves optimal convergence rates in the stochastic setting, and ensures simultaneous coverage and efficiency relative to dynamic oracle thresholds under covariate-dependent shifts.
0 citationsRead paper
Deterministic Algorithms for Low Individual Degree Factors of Sparse Polynomials
This work addresses the problem of efficiently and deterministically enumerating all factors of bounded individual degree in sparse polynomials. For an $n$-variate, $s$-sparse polynomial with individual degrees at most $d$, the paper presents the first deterministic quasi-polynomial time algorithm that works over arbitrary fields and accommodates unbounded total degree. The approach combines interpolation, divisibility testing, and algebraic complexity techniques to construct constant-depth arithmetic circuits capturing all such factors, achieving a running time of $\mathrm{poly}(n, s^d)$. For general sparse polynomials, the algorithm runs in time $\mathrm{poly}(D^{d \log s}, s^{d^2 \log n})$, where $D$ bounds the total degree. These results generalize prior work and yield new upper bounds on the number of low individual-degree factors.
0 citationsRead paper