LIN LI

LIN LI


Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
4 October 2025 in Numerical Algorithms
Verified
Journal Article
Open access
PDF file
LLLIN LI
LIN LI
JHJames M Hyman
James M Hyman
We introduce a control variate integration (CVI) method using radial basis functions (RBFs) for high-dimensional numerical integration with sparse sample data. Unlike polynomial-based CVI, which captures large-scale variations effectively but suffers from exponential basis function growth, our RBF approach adapts to both large- and small-scale data variations. For problems with moderately high dimensions ( $$d \le 20$$ ), numerical experiments demonstrate that the RBF-based CVI significantly outperforms the traditional Monte Carlo and Quasi-Monte Carlo methods. Although the polynomial CVI typically preserves the convergence rate of the underlying sampling method, our RBF CVI demonstrates substantially faster convergence rates. Empirical observations show convergence rates between $$\mathcal {O}(N^{-1.5})$$ and $$\mathcal {O}(N^{-2})$$ in multiple test cases. This performance advantage holds for both independent and low-discrepancy sampling scenarios, offering a practical solution for high-dimensional integration problems with sparse data commonly encountered in computational finance, Bayesian inference, and uncertainty quantification.

7 May 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
LLLIN LI
LIN LI
CZChaochao Zhou
Chaochao Zhou
BABenjamin Aubert
Benjamin Aubert
JGJunlin Guo
Junlin Guo
JZJunchao Zhu
Junchao Zhu
Rigid 2D-3D image registration plays a critical role in modern image-guided interventions by enabling the alignment of intraoperative X-ray images with preoperative volumetric data such as computed tomography (CT). Accurate 2D-3D registration allows clinicians to localize anatomical structures in three-dimensional space while relying on fast and low-dose intraoperative imaging, which is essential for applications including orthopedic surgery, spine navigation, radiation therapy, and interventional radiology. Despite significant progress over the past two decades, achieving robust and accurate registration remains challenging due to factors such as limited imaging viewpoints, occlusions, imaging noise, and the large search space of rigid transformations. This paper provides a comprehensive survey of rigid 2D-3D registration methods with a particular focus on X-ray-to-CT alignment. We first introduce the mathematical formulation of the registration problem and present a taxonomy of existing approaches. We then review the key technical components underlying modern registration systems. In addition, we summarize commonly used datasets and evaluation protocols, discussing widely adopted metrics such as target registration error (TRE), pose error, and reprojection error. The survey also highlights representative clinical applications and analyzes the practical challenges that remain in real-world deployment, including robustness to imaging artifacts, variations in imaging dose, and real-time computational constraints. Finally, we discuss emerging research directions, such as differentiable rendering, deep learning based pose estimation, and multi-view registration frameworks, which are expected to further improve the accuracy, robustness, and clinical applicability of 2D-3D registration methods.

6 July 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
LLLIN LI
LIN LI
CZChaochao Zhou
Chaochao Zhou
Reinforcement learning (RL) has become an important component of modern artificial intelligence research and practice. It underlies systems that achieved strong performance at Go, enabled robot locomotion, and has been applied in the training pipelines of large language models. Yet the primary literature remains fragmented: scattered across thousands of specialist papers, dense with notation, and rarely synthesised into a narrative accessible to practitioners or researchers from adjacent fields. This paper presents an expository survey that prioritises conceptual clarity over formal rigour. Rather than cataloguing algorithms through convergence proofs or sample-complexity bounds, we aim to build understanding from the ground up, using intuition, analogy, and high-level pseudocode as primary expository tools. We acknowledge that this approach involves deliberate trade-offs: informal treatments can obscure edge cases, and readers transitioning to the primary literature will need to supplement this survey with more rigorous references. Our coverage is deliberately broad: we trace the conceptual arc from classical tabular methods—dynamic programming, Monte Carlo estimation, temporal-difference learning—through the deep RL revolution, to frontier topics including offline RL, reinforcement learning from human feedback (RLHF), preference optimisation, and model-based planning. The paper is structured as a self-contained introduction. A reader who has never encountered a Bellman equation will find a plain-language introduction; a practitioner familiar with Q-learning will find a unified framework that situates that knowledge within the broader landscape and connects it to methods such as PPO, CQL, DPO, and MuZero. Each algorithm is introduced as an intuition first, then sketched in readable pseudocode with at least one grounding analogy. The intended contribution is an accessible introductory reference—a single, coherent document through which a motivated non-specialist can acquire the vocabulary, conceptual map, and practical orientation needed to engage with the primary literature. It is not a replacement for rigorous treatments such as [1] or [2], to which the reader is directed for formal foundations.

25 June 2025 in Research Square
Verified
Preprint
Open access
PDF file
LLLIN LI
LIN LI
JHJames M Hyman
James M Hyman
We introduce a control variate integration (CVI) method using radial basis functions (RBFs) for high-dimensional numerical integration with sparse sample data. Unlike polynomial-based CVI, which effectively captures large-scale variations but suffers from the curse of dimensionality, our RBF approach inherently captures both large- and small-scale data variations. For problems with moderately high dimensions (d ≤ 20), numerical experiments demonstrate that the RBF-based CVI significantly outperforms the traditional Monte Carlo and Quasi-Monte Carlo methods. Although the polynomial CVI typically maintains the same convergence rate as the underlying sampling method, our RBF CVI achieves substantially faster convergence rates, empirically observed between O(N ^{−1.5}) and O(N ^{−2}). This performance advantage holds for both independent and low-discrepancy sampling scenarios, offering a practical solution for high-dimensional integration problems with sparse data.