Progressive Pseudo-Label Optimization for Point-Supervised Change Detection
该研究针对点监督变化检测中伪标签不完整和噪声问题，提出两阶段框架结合SAM2先验逐步优化伪标签，提高边界质量和结构一致性。
0 citationsRead paper
Selected work
Representative Papers
Efficient Coreset Selection via K-Nearest Neighbor Graphs
This study addresses the prohibitive spatiotemporal overhead of existing coreset selection methods on large-scale data by proposing knng-cs, a lightweight approach based on k-nearest neighbor graphs. By leveraging local neighborhood structures to estimate data importance and greedily selecting representative nodes, this method eliminates the need for maintaining dense distance matrices, thereby achieving linear storage complexity and efficient filtering. Experiments across four real-world datasets demonstrate that knng-cs achieves accuracy comparable to state-of-the-art baselines while reducing selection time by 2.3–41.2× and peak memory usage to merely 0.3%–7.5% of baseline levels. These results indicate that knng-cs significantly enhances both the efficiency and scalability of large-scale coreset construction without compromising representational quality.
0 citationsRead paper
Efficient Privacy-Preserving Range Filtered Approximate Nearest Neighbor Search
This study addresses privacy leakage in outsourced vector database range-filtering approximate nearest neighbor search (RF-ANNS) by proposing the first privacy-preserving scheme operating within an encrypted environment. By decoupling range localization from encrypted search, the authors construct an efficient filtering-refinement pipeline integrating n-ary attribute trees, neighbor graph sub-indexes, and distance-comparison-preserving encryption. Experimental evaluations across four datasets demonstrate that this protocol achieves a significantly superior query-per-second versus recall trade-off compared to existing secure methods while maintaining robust scalability. This work represents the first systematic implementation and validation of secure, high-performance retrieval for RF-ANNS in outsourced encrypted vector databases, effectively resolving critical privacy concerns without compromising search efficiency.
0 citationsRead paper
FROG: Efficient Range-Filtering Approximate Nearest Neighbor Search on GPUs
This study addresses the low throughput, poor scalability, and redundant computation inherent in existing GPU-based filtered approximate nearest neighbor search. To overcome these limitations, we propose FROG, a novel indexing framework that replaces conventional local subgraph optimization with a global-aware vertex-centric architecture. By constructing GPU-friendly neighbor candidate structures and designing specialized algorithms, FROG enables efficient index construction and rapid neighbor expansion. Experimental evaluations demonstrate that FROG significantly outperforms state-of-the-art baselines, achieving 14.7–37.7× higher throughput than CPU implementations and 4.5–7.6× improvement over the best GPU alternatives on mixed-selectivity queries. Furthermore, it accelerates index building by 2.4–14.8×, effectively breaking through current performance bottlenecks in high-dimensional filtered retrieval tasks.
0 citationsRead paper
Identifying parameter couplings and uncertainties of mixed-noise stochastic systems via full-covariance Gaussian mixture network
This study addresses the challenge of parameter identification in stochastic systems with mixed noise, where intractable likelihoods hinder conventional estimation. We propose Penn-GMD, a network that maps trajectories to full-covariance Gaussian mixture distributions and employs surjective parameterization with negative log-likelihood minimization to approximate the true likelihood. The core contribution lies in leveraging full covariance matrices to explicitly reveal parameter coupling and multimodal structures. This approach not only accurately recovers the underlying likelihood distribution but also naturally diagnoses unidentifiability. Consequently, Penn-GMD effectively resolves persistent difficulties in parameter estimation and uncertainty quantification for complex stochastic systems where traditional methods typically fail, offering a robust framework for handling intractable inference problems in noisy dynamical environments.
0 citationsRead paper
Recent publications
Latest Papers
Progressive Pseudo-Label Optimization for Point-Supervised Change Detection
该研究针对点监督变化检测中伪标签不完整和噪声问题，提出两阶段框架结合SAM2先验逐步优化伪标签，提高边界质量和结构一致性。
0 citationsRead paper
Efficient Coreset Selection via K-Nearest Neighbor Graphs
This study addresses the prohibitive spatiotemporal overhead of existing coreset selection methods on large-scale data by proposing knng-cs, a lightweight approach based on k-nearest neighbor graphs. By leveraging local neighborhood structures to estimate data importance and greedily selecting representative nodes, this method eliminates the need for maintaining dense distance matrices, thereby achieving linear storage complexity and efficient filtering. Experiments across four real-world datasets demonstrate that knng-cs achieves accuracy comparable to state-of-the-art baselines while reducing selection time by 2.3–41.2× and peak memory usage to merely 0.3%–7.5% of baseline levels. These results indicate that knng-cs significantly enhances both the efficiency and scalability of large-scale coreset construction without compromising representational quality.
0 citationsRead paper
Efficient Privacy-Preserving Range Filtered Approximate Nearest Neighbor Search
This study addresses privacy leakage in outsourced vector database range-filtering approximate nearest neighbor search (RF-ANNS) by proposing the first privacy-preserving scheme operating within an encrypted environment. By decoupling range localization from encrypted search, the authors construct an efficient filtering-refinement pipeline integrating n-ary attribute trees, neighbor graph sub-indexes, and distance-comparison-preserving encryption. Experimental evaluations across four datasets demonstrate that this protocol achieves a significantly superior query-per-second versus recall trade-off compared to existing secure methods while maintaining robust scalability. This work represents the first systematic implementation and validation of secure, high-performance retrieval for RF-ANNS in outsourced encrypted vector databases, effectively resolving critical privacy concerns without compromising search efficiency.
0 citationsRead paper
FROG: Efficient Range-Filtering Approximate Nearest Neighbor Search on GPUs
This study addresses the low throughput, poor scalability, and redundant computation inherent in existing GPU-based filtered approximate nearest neighbor search. To overcome these limitations, we propose FROG, a novel indexing framework that replaces conventional local subgraph optimization with a global-aware vertex-centric architecture. By constructing GPU-friendly neighbor candidate structures and designing specialized algorithms, FROG enables efficient index construction and rapid neighbor expansion. Experimental evaluations demonstrate that FROG significantly outperforms state-of-the-art baselines, achieving 14.7–37.7× higher throughput than CPU implementations and 4.5–7.6× improvement over the best GPU alternatives on mixed-selectivity queries. Furthermore, it accelerates index building by 2.4–14.8×, effectively breaking through current performance bottlenecks in high-dimensional filtered retrieval tasks.
0 citationsRead paper
Identifying parameter couplings and uncertainties of mixed-noise stochastic systems via full-covariance Gaussian mixture network
This study addresses the challenge of parameter identification in stochastic systems with mixed noise, where intractable likelihoods hinder conventional estimation. We propose Penn-GMD, a network that maps trajectories to full-covariance Gaussian mixture distributions and employs surjective parameterization with negative log-likelihood minimization to approximate the true likelihood. The core contribution lies in leveraging full covariance matrices to explicitly reveal parameter coupling and multimodal structures. This approach not only accurately recovers the underlying likelihood distribution but also naturally diagnoses unidentifiability. Consequently, Penn-GMD effectively resolves persistent difficulties in parameter estimation and uncertainty quantification for complex stochastic systems where traditional methods typically fail, offering a robust framework for handling intractable inference problems in noisy dynamical environments.
0 citationsRead paper