Robust Policy Optimization via Adversarial Importance Sampling
本文通过引入对抗重要性采样方法（Advis）来解决深度强化学习政策对输入扰动的鲁棒性问题，无需额外环境交互或辅助网络，并提供了一个PyTorch库支持快速原型设计与评估。
0 citationsRead paper
Selected work
Representative Papers
A Unified and Constrained View of Regularization-Based Robust Reinforcement Learning
本文通过统一正则化方法并导出性能差距的上界，提出一种约束优化问题来提高深度强化学习策略在对抗性输入扰动下的鲁棒性。
0 citationsRead paper
3D Euler-Angle Orientation Control for Two-Ray Fading Mitigation in Maritime Air-to-Sea Communications
本文通过控制无人机的三维姿态来优化海上空对海通信中的两射线干扰问题，使用非线性模型预测控制方法，提高了通信吞吐量。
0 citationsRead paper
Bulbul: A Dataset for Dialectal Arabic Speech Recognition
为解决阿拉伯语自动语音识别面临的独特挑战，通过构建包含275名来自11个阿拉伯国家的说话者的多方言数据集BULBUL，并采用两层人工验证确保质量。
0 citationsRead paper
Extracting and Verifying Illicit Bitcoin Addresses from Underground Forum Discussions
This study addresses the irreproducibility and lack of direct evidence in existing Bitcoin illicit labeling datasets by proposing an evidence-driven label construction paradigm. Integrating large language model screening, expert review, and on-chain transaction verification, we establish a fully transparent pipeline for extracting and validating illicit addresses from underground forums. We release a comprehensive dataset comprising 2,438 manually verified illicit addresses alongside complete code tools. This contribution effectively resolves the challenge of missing evidentiary support, significantly enhancing both dataset credibility and research reproducibility in cryptocurrency forensics.
0 citationsRead paper
Recent publications
Latest Papers
Robust Policy Optimization via Adversarial Importance Sampling
本文通过引入对抗重要性采样方法（Advis）来解决深度强化学习政策对输入扰动的鲁棒性问题，无需额外环境交互或辅助网络，并提供了一个PyTorch库支持快速原型设计与评估。
0 citationsRead paper
A Unified and Constrained View of Regularization-Based Robust Reinforcement Learning
本文通过统一正则化方法并导出性能差距的上界，提出一种约束优化问题来提高深度强化学习策略在对抗性输入扰动下的鲁棒性。
0 citationsRead paper
3D Euler-Angle Orientation Control for Two-Ray Fading Mitigation in Maritime Air-to-Sea Communications
本文通过控制无人机的三维姿态来优化海上空对海通信中的两射线干扰问题，使用非线性模型预测控制方法，提高了通信吞吐量。
0 citationsRead paper
Bulbul: A Dataset for Dialectal Arabic Speech Recognition
为解决阿拉伯语自动语音识别面临的独特挑战，通过构建包含275名来自11个阿拉伯国家的说话者的多方言数据集BULBUL，并采用两层人工验证确保质量。
0 citationsRead paper
Extracting and Verifying Illicit Bitcoin Addresses from Underground Forum Discussions
This study addresses the irreproducibility and lack of direct evidence in existing Bitcoin illicit labeling datasets by proposing an evidence-driven label construction paradigm. Integrating large language model screening, expert review, and on-chain transaction verification, we establish a fully transparent pipeline for extracting and validating illicit addresses from underground forums. We release a comprehensive dataset comprising 2,438 manually verified illicit addresses alongside complete code tools. This contribution effectively resolves the challenge of missing evidentiary support, significantly enhancing both dataset credibility and research reproducibility in cryptocurrency forensics.
0 citationsRead paper