Selected work

Representative Papers

Beyond Background Bias: Saliency-Driven Prototype Alignment for Dataset Distillation

Jul 28, 2026

Existing diffusion-based dataset distillation methods struggle to balance structural consistency and generalization due to weak alignment between latent prototypes and class-discriminative regions, as well as susceptibility to background interference. This work proposes a saliency-driven two-stage prototype alignment framework that operates without fine-tuning the frozen diffusion backbone (e.g., LDM or DiT). By integrating Grad-CAM to generate high-confidence discriminative regions and introducing a hard prototype refinement strategy to enhance prototype diversity and discriminability, the method leverages only a lightweight classifier to achieve significant improvements over strong baselines across multiple benchmarks. The approach effectively boosts the representativeness, training efficacy, and generalization capability of synthesized data.

0 citationsRead paper

SLAM: Structured and Localized Analytic Manifold Adaptation for Lifelong VPR

Jul 06, 2026

This work addresses the challenge of lifelong visual place recognition (VPR), where systems must continuously adapt to novel environments while avoiding catastrophic forgetting. To this end, the authors propose a structured, localized analytical manifold adaptation framework that uniquely integrates uncertainty-aware smoothing—based on the unscented transform—Gaussian mixture model–driven topological space partitioning, and H∞ robust bound optimization into a unified closed-form recursive formulation. This integration enables precise control over the trade-off between accuracy and robustness through a single regularization parameter. Experimental results demonstrate that the U+G configuration achieves a state-of-the-art nominal accuracy of 27.5%, while the full H∞ deployment delivers minimax robust performance with formal mathematical guarantees, all without requiring architectural decomposition.

0 citationsRead paper

FlatManifold: Robust Continual Learning under Severe Label Noise and Domain Shifts via Intrinsic Manifold Flattening

Jul 06, 2026

This work addresses the dual challenges of severe uncalibrated label noise and complex nonlinear domain shifts in non-stationary streaming scenarios by proposing a robust continual learning framework based on Nyström manifold flattening mappings. Departing from conventional sample filtering strategies, the method leverages kernel tricks to project feature distributions into an orthogonalized reproducing kernel Hilbert space (RKHS), integrating ridge regularization with a covariance-based topological braking term to structurally suppress label noise and mitigate catastrophic forgetting during optimization. Evaluated on real-world robotic multi-session data featuring 40% symmetric label noise and drastic cross-seasonal illumination changes, the approach significantly outperforms existing baselines, effectively alleviating gradient contamination while achieving high generalization performance.

0 citationsRead paper

Understanding Geometric Representations in Self-Supervised Vision Transformers via Subspace Intervention

Jul 02, 2026

Existing linear probes struggle to uncover the internal encoding structure of geometric information in self-supervised vision Transformers (ViTs). This work proposes a controlled subspace intervention framework that leverages singular value decomposition (SVD) on converged linear probe weights to isolate a low-rank subspace carrying explicit geometric signals. For the first time, subspace analysis reveals distinct differences in geometric representation between DINOv2 and MAE, demonstrating that geometric information is highly compressible, peaks in accuracy at intermediate network layers, and exhibits pronounced low-rank characteristics. These findings provide both theoretical grounding and practical design guidance for lightweight decoders and efficient feature selection strategies in self-supervised vision models.

0 citationsRead paper

How Can Size and Ceiling Bounds Affect the Complexity of Nonuniform Automata Families?

Jun 25, 2026

This study investigates the impact of state size and input-length ceiling on the computational power of non-uniform families of finite automata and pushdown automata. By integrating formal language theory, automata theory, and space-bounded complexity theory with the Karp–Lipton advice mechanism, the work systematically analyzes the individual and joint effects of these two parameters. It is the first to explicitly distinguish and quantitatively characterize their respective roles, establishing a precise connection between non-uniform automaton families and space-bounded complexity classes augmented with length-bounded advice. This advances the understanding of non-uniform computational models and offers a novel perspective for relating classical complexity classes.

0 citationsRead paper

Recent publications

Latest Papers

Beyond Background Bias: Saliency-Driven Prototype Alignment for Dataset Distillation

Jul 28, 2026

Existing diffusion-based dataset distillation methods struggle to balance structural consistency and generalization due to weak alignment between latent prototypes and class-discriminative regions, as well as susceptibility to background interference. This work proposes a saliency-driven two-stage prototype alignment framework that operates without fine-tuning the frozen diffusion backbone (e.g., LDM or DiT). By integrating Grad-CAM to generate high-confidence discriminative regions and introducing a hard prototype refinement strategy to enhance prototype diversity and discriminability, the method leverages only a lightweight classifier to achieve significant improvements over strong baselines across multiple benchmarks. The approach effectively boosts the representativeness, training efficacy, and generalization capability of synthesized data.

0 citationsRead paper

SLAM: Structured and Localized Analytic Manifold Adaptation for Lifelong VPR

Jul 06, 2026

This work addresses the challenge of lifelong visual place recognition (VPR), where systems must continuously adapt to novel environments while avoiding catastrophic forgetting. To this end, the authors propose a structured, localized analytical manifold adaptation framework that uniquely integrates uncertainty-aware smoothing—based on the unscented transform—Gaussian mixture model–driven topological space partitioning, and H∞ robust bound optimization into a unified closed-form recursive formulation. This integration enables precise control over the trade-off between accuracy and robustness through a single regularization parameter. Experimental results demonstrate that the U+G configuration achieves a state-of-the-art nominal accuracy of 27.5%, while the full H∞ deployment delivers minimax robust performance with formal mathematical guarantees, all without requiring architectural decomposition.

0 citationsRead paper

FlatManifold: Robust Continual Learning under Severe Label Noise and Domain Shifts via Intrinsic Manifold Flattening

Jul 06, 2026

This work addresses the dual challenges of severe uncalibrated label noise and complex nonlinear domain shifts in non-stationary streaming scenarios by proposing a robust continual learning framework based on Nyström manifold flattening mappings. Departing from conventional sample filtering strategies, the method leverages kernel tricks to project feature distributions into an orthogonalized reproducing kernel Hilbert space (RKHS), integrating ridge regularization with a covariance-based topological braking term to structurally suppress label noise and mitigate catastrophic forgetting during optimization. Evaluated on real-world robotic multi-session data featuring 40% symmetric label noise and drastic cross-seasonal illumination changes, the approach significantly outperforms existing baselines, effectively alleviating gradient contamination while achieving high generalization performance.

0 citationsRead paper

Understanding Geometric Representations in Self-Supervised Vision Transformers via Subspace Intervention

Jul 02, 2026

Existing linear probes struggle to uncover the internal encoding structure of geometric information in self-supervised vision Transformers (ViTs). This work proposes a controlled subspace intervention framework that leverages singular value decomposition (SVD) on converged linear probe weights to isolate a low-rank subspace carrying explicit geometric signals. For the first time, subspace analysis reveals distinct differences in geometric representation between DINOv2 and MAE, demonstrating that geometric information is highly compressible, peaks in accuracy at intermediate network layers, and exhibits pronounced low-rank characteristics. These findings provide both theoretical grounding and practical design guidance for lightweight decoders and efficient feature selection strategies in self-supervised vision models.

0 citationsRead paper

How Can Size and Ceiling Bounds Affect the Complexity of Nonuniform Automata Families?

Jun 25, 2026

This study investigates the impact of state size and input-length ceiling on the computational power of non-uniform families of finite automata and pushdown automata. By integrating formal language theory, automata theory, and space-bounded complexity theory with the Karp–Lipton advice mechanism, the work systematically analyzes the individual and joint effects of these two parameters. It is the first to explicitly distinguish and quantitatively characterize their respective roles, establishing a precise connection between non-uniform automaton families and space-bounded complexity classes augmented with length-bounded advice. This advances the understanding of non-uniform computational models and offers a novel perspective for relating classical complexity classes.

0 citationsRead paper