Selected work

Representative Papers

Persistent Homology-Guided Frequency Filtering for Image Compression

Dec 07, 2025

To address semantic information loss and low feature reliability in noisy image compression, this paper proposes a persistent homology-guided frequency-domain filtering method. First, discrete Fourier transform (DFT) is applied to input images; then, persistent homology analysis identifies topologically significant frequency components critical for classification tasks; finally, structure-preserving filtering is performed in the frequency domain to achieve noise-robust compression and reconstruction. This work is the first to embed topological data analysis into the image frequency-domain compression pipeline, explicitly preserving semantic-relevant topological structures. Evaluated on CNN-based downstream binary classification tasks, the method matches JPEG’s performance across six compression quality metrics—including PSNR and SSIM—while significantly enhancing feature discriminability and classification accuracy on noisy images. The approach establishes a novel paradigm for robust image representation learning.

0 citationsRead paper

Atmospheric model-trained machine learning selection and classification of ultracool TY dwarfs

Jul 01, 2025

Current brown dwarf surveys suffer from severe incompleteness in detecting late-type T/Y dwarfs (spectral types T8 and later), primarily due to the scarcity of empirically confirmed objects. Method: We present the first machine learning classification framework trained exclusively on synthetic photometric data generated from state-of-the-art atmospheric models (ATMO 2020 and Sonora Bobcat), enabling construction of a high-fidelity training set two orders of magnitude larger than existing observational samples. Spectral types are automatically assigned via polynomial color–spectral type relations, and classification is performed using an ensemble learner. Contribution/Results: The framework achieves >99% classification accuracy on both synthetic and real-world datasets, with spectral type predictions accurate to 0.35±0.37 subclasses. It successfully identified an uncatalogued T8.2 candidate in the Pisces field and the UKIDSS Ultra-Deep Survey (UDS) region, substantially enhancing the completeness and efficiency of ultracool dwarf surveys.

0 citationsRead paper

Redefining Network Topology in Complex Systems: Merging Centrality Metrics, Spectral Theory, and Diffusion Dynamics

Mar 27, 2025

Static centrality measures fail to capture perturbation propagation dynamics and resilience bottlenecks in complex network analysis. Method: This paper proposes a unified modeling framework integrating node centrality, graph Laplacian spectral features, and continuous-time diffusion dynamics. It enables the first synergistic modeling and multi-dimensional joint optimization of these three indicator classes, overcoming limitations of unidimensional assessment. Leveraging spectral analysis and stochastic diffusion processes, we develop an interpretable method for critical node identification and vulnerability localization. Results: Validation on synthetic networks demonstrates a 23.6% improvement in critical node identification accuracy, significantly enhancing detection of propagation bottlenecks and robustness weaknesses. The framework provides theoretical foundations and decision-support tools for epidemic control and cybersecurity hardening.

0 citationsRead paper

Efficient Knowledge Feeding to Language Models: A Novel Integrated Encoder-Decoder Architecture

Feb 07, 2025

This work addresses two key limitations: (1) retrieval-augmented generation (RAG) suffers from token-length constraints and retrieval inaccuracy, and (2) in-context learning (ICL) relies heavily on numerous annotated examples. To overcome these, we propose the In-Context Vector (ICV) mechanism—a task-aware, encoder-decoder-based approach that encodes external knowledge into compact latent vectors and directly modulates the internal hidden states of large language models (LLMs), enabling example-free, low-overhead knowledge injection. Crucially, ICV replaces explicit prompts with implicit vector representations, eliminating dependence on external retrieval systems and lengthy prompts. Experiments demonstrate that ICV significantly outperforms standard ICL and fine-tuning baselines on question answering and information retrieval tasks. It achieves competitive performance with minimal parameter overhead—orders of magnitude smaller than LLaMA-3, Gemma, or Phi-3—and substantially reduces computational cost, memory footprint, and input sequence length.

0 citationsRead paper

Recent publications

Latest Papers

Persistent Homology-Guided Frequency Filtering for Image Compression

Dec 07, 2025

To address semantic information loss and low feature reliability in noisy image compression, this paper proposes a persistent homology-guided frequency-domain filtering method. First, discrete Fourier transform (DFT) is applied to input images; then, persistent homology analysis identifies topologically significant frequency components critical for classification tasks; finally, structure-preserving filtering is performed in the frequency domain to achieve noise-robust compression and reconstruction. This work is the first to embed topological data analysis into the image frequency-domain compression pipeline, explicitly preserving semantic-relevant topological structures. Evaluated on CNN-based downstream binary classification tasks, the method matches JPEG’s performance across six compression quality metrics—including PSNR and SSIM—while significantly enhancing feature discriminability and classification accuracy on noisy images. The approach establishes a novel paradigm for robust image representation learning.

0 citationsRead paper

Atmospheric model-trained machine learning selection and classification of ultracool TY dwarfs

Jul 01, 2025

Current brown dwarf surveys suffer from severe incompleteness in detecting late-type T/Y dwarfs (spectral types T8 and later), primarily due to the scarcity of empirically confirmed objects. Method: We present the first machine learning classification framework trained exclusively on synthetic photometric data generated from state-of-the-art atmospheric models (ATMO 2020 and Sonora Bobcat), enabling construction of a high-fidelity training set two orders of magnitude larger than existing observational samples. Spectral types are automatically assigned via polynomial color–spectral type relations, and classification is performed using an ensemble learner. Contribution/Results: The framework achieves >99% classification accuracy on both synthetic and real-world datasets, with spectral type predictions accurate to 0.35±0.37 subclasses. It successfully identified an uncatalogued T8.2 candidate in the Pisces field and the UKIDSS Ultra-Deep Survey (UDS) region, substantially enhancing the completeness and efficiency of ultracool dwarf surveys.

0 citationsRead paper

Redefining Network Topology in Complex Systems: Merging Centrality Metrics, Spectral Theory, and Diffusion Dynamics

Mar 27, 2025

Static centrality measures fail to capture perturbation propagation dynamics and resilience bottlenecks in complex network analysis. Method: This paper proposes a unified modeling framework integrating node centrality, graph Laplacian spectral features, and continuous-time diffusion dynamics. It enables the first synergistic modeling and multi-dimensional joint optimization of these three indicator classes, overcoming limitations of unidimensional assessment. Leveraging spectral analysis and stochastic diffusion processes, we develop an interpretable method for critical node identification and vulnerability localization. Results: Validation on synthetic networks demonstrates a 23.6% improvement in critical node identification accuracy, significantly enhancing detection of propagation bottlenecks and robustness weaknesses. The framework provides theoretical foundations and decision-support tools for epidemic control and cybersecurity hardening.

0 citationsRead paper

Efficient Knowledge Feeding to Language Models: A Novel Integrated Encoder-Decoder Architecture

Feb 07, 2025

This work addresses two key limitations: (1) retrieval-augmented generation (RAG) suffers from token-length constraints and retrieval inaccuracy, and (2) in-context learning (ICL) relies heavily on numerous annotated examples. To overcome these, we propose the In-Context Vector (ICV) mechanism—a task-aware, encoder-decoder-based approach that encodes external knowledge into compact latent vectors and directly modulates the internal hidden states of large language models (LLMs), enabling example-free, low-overhead knowledge injection. Crucially, ICV replaces explicit prompts with implicit vector representations, eliminating dependence on external retrieval systems and lengthy prompts. Experiments demonstrate that ICV significantly outperforms standard ICL and fine-tuning baselines on question answering and information retrieval tasks. It achieves competitive performance with minimal parameter overhead—orders of magnitude smaller than LLaMA-3, Gemma, or Phi-3—and substantially reduces computational cost, memory footprint, and input sequence length.

0 citationsRead paper