The Unbearable Weight: Scaling Models and Methods for UAV Audio Classification
研究解决了无人机音频分类中资源受限的问题，通过对比不同模型架构和微调方法，发现轻量级CNN在准确性和效率上优于大型预训练模型。
0 citationsRead paper
Selected work
Representative Papers
Group Isomorphism and the Polylogarithmic-Time Hierarchy: Depth-2$\frac{1}{2}$ Circuits and Lower Bounds
本文研究了群同构问题的低深度电路复杂性，通过一种基于群结构的新策略，证明了深度2布尔电路需要准多项式大小，并提供了深度2.5电路的上界。
0 citationsRead paper
Students' Perceptions of Peer Grading
研究通过分析文献和Reddit讨论，探讨学生对同伴评分的看法及其利弊，并提出缓解策略，特别是针对AI使用可能引发的信任问题。
0 citationsRead paper
Parallel Algorithms for Group Isomorphism via Code Equivalence
This work addresses the isomorphism testing problem for coprime extension groups and central-radical groups, aiming to achieve efficient parallel algorithms. By introducing, for the first time, parallelization techniques from linear code equivalence into group isomorphism testing, and integrating Luks’s group-theoretic framework with the AC circuit model, the authors exploit structural properties of group multiplication tables for optimization. The main contribution lies in achieving isomorphism testing for these two classes of groups within the complexity class AC³. Furthermore, the circuit depth for isomorphism testing of arbitrary central-radical groups is reduced to O(log³n) with circuit size n^{O(log log n)}, significantly improving upon prior results.
0 citationsRead paper
A Multiclass Acoustic Dataset and Interactive Tool for Analyzing Drone Signatures in Real-World Environments
Existing UAV detection systems exhibit limited performance in challenging environments—such as low-light conditions and long-range scenarios—where visual and RF-based methods often fail. Method: This work proposes an acoustic-feature-based multi-class UAV identification framework. We introduce the first publicly available, high-quality acoustic dataset comprising 32 mainstream UAV models, including raw audio recordings, time-frequency spectrograms, and MFCC feature maps, all captured under realistic multi-distance and multi-angle conditions. Concurrently, we develop a lightweight, interactive web platform enabling real-time audio playback, spectrogram visualization, and feature exploration. Contribution/Results: The dataset significantly enhances discriminability of acoustic signatures for UAV classification. Empirical evaluation demonstrates strong generalization capability across diverse UAV models and operating conditions. The platform has received positive feedback from researchers and educators. This work provides foundational data resources and open-source tools to advance UAV acoustic sensing, edge-based detection, and AI-enabled security governance.
0 citationsRead paper
Recent publications
Latest Papers
The Unbearable Weight: Scaling Models and Methods for UAV Audio Classification
研究解决了无人机音频分类中资源受限的问题，通过对比不同模型架构和微调方法，发现轻量级CNN在准确性和效率上优于大型预训练模型。
0 citationsRead paper
Group Isomorphism and the Polylogarithmic-Time Hierarchy: Depth-2$\frac{1}{2}$ Circuits and Lower Bounds
本文研究了群同构问题的低深度电路复杂性，通过一种基于群结构的新策略，证明了深度2布尔电路需要准多项式大小，并提供了深度2.5电路的上界。
0 citationsRead paper
Students' Perceptions of Peer Grading
研究通过分析文献和Reddit讨论，探讨学生对同伴评分的看法及其利弊，并提出缓解策略，特别是针对AI使用可能引发的信任问题。
0 citationsRead paper
Parallel Algorithms for Group Isomorphism via Code Equivalence
This work addresses the isomorphism testing problem for coprime extension groups and central-radical groups, aiming to achieve efficient parallel algorithms. By introducing, for the first time, parallelization techniques from linear code equivalence into group isomorphism testing, and integrating Luks’s group-theoretic framework with the AC circuit model, the authors exploit structural properties of group multiplication tables for optimization. The main contribution lies in achieving isomorphism testing for these two classes of groups within the complexity class AC³. Furthermore, the circuit depth for isomorphism testing of arbitrary central-radical groups is reduced to O(log³n) with circuit size n^{O(log log n)}, significantly improving upon prior results.
0 citationsRead paper
A Multiclass Acoustic Dataset and Interactive Tool for Analyzing Drone Signatures in Real-World Environments
Existing UAV detection systems exhibit limited performance in challenging environments—such as low-light conditions and long-range scenarios—where visual and RF-based methods often fail. Method: This work proposes an acoustic-feature-based multi-class UAV identification framework. We introduce the first publicly available, high-quality acoustic dataset comprising 32 mainstream UAV models, including raw audio recordings, time-frequency spectrograms, and MFCC feature maps, all captured under realistic multi-distance and multi-angle conditions. Concurrently, we develop a lightweight, interactive web platform enabling real-time audio playback, spectrogram visualization, and feature exploration. Contribution/Results: The dataset significantly enhances discriminability of acoustic signatures for UAV classification. Empirical evaluation demonstrates strong generalization capability across diverse UAV models and operating conditions. The platform has received positive feedback from researchers and educators. This work provides foundational data resources and open-source tools to advance UAV acoustic sensing, edge-based detection, and AI-enabled security governance.
0 citationsRead paper