Selected work

Representative Papers

On the Equivalence of Optimal Transport Problem and Action Matching with Optimal Vector Fields

Oct 31, 2025

Mapping between sequences of continuous probability distributions arises in generative modeling, yet existing optimal transport (OT) methods suffer from high computational cost due to explicit joint distribution estimation or Wasserstein distance computation. Method: We establish a rigorous equivalence between Action Matching (AM) and OT by formulating distributional transport as learning an optimal vector field for a generative ordinary differential equation (ODE). Under compatible boundary conditions and energy functional design, the AM solution inherently satisfies the Kantorovich duality optimality criterion. Contribution/Results: This equivalence reveals that OT’s optimal coupling is fully characterized by a specific class of integrable vector fields—bypassing explicit joint distribution optimization or Wasserstein metric evaluation. Empirically, AM preserves theoretical optimality while achieving substantial computational efficiency gains. Our work provides a novel theoretical foundation and optimization paradigm for flow-matching-based generative modeling.

0 citationsRead paper

Comparison of Unsupervised Metrics for Evaluating Judicial Decision Extraction

Oct 02, 2025

Legal NLP lacks human-annotated data for evaluating semantic block extraction quality in judicial judgments. Method: We systematically benchmark 16 unsupervised metrics across seven legal content extraction tasks, integrating document-level, semantic-level, structural-level, pseudo-labeling, and law-specific measures—including TF-Coherence, Coverage Ratio, Legal Term Density, and LLM-based scoring—validated via bootstrap correlation analysis, Lin’s concordance correlation coefficient, and mean absolute error (MAE) against expert judgments. Results: TF-Coherence and Coverage Ratio achieve the strongest alignment with expert ratings (r > 0.5, MAE < 0.14), significantly outperforming LLM scoring (r = 0.382), thereby exposing LLMs’ limitations in fine-grained legal assessment. This work introduces the first scalable, unsupervised evaluation framework tailored to legal text, balancing computational efficiency and reliability, and enabling annotation-free, automated quality screening for large-scale legal NLP systems.

0 citationsRead paper

Error correction in multiclass image classification of facial emotion on unbalanced samples

Oct 02, 2025

To address recognition bias toward minority classes in facial emotion multi-class classification caused by class imbalance, this paper proposes a deep network integrating LSTM with spatial attention mechanisms to model emotion features from discriminative facial regions. We introduce a novel six-class subset training strategy and incorporate a seventh class—comprising external rare emotion instances—to enable cross-class error correction. Extensive experiments across multiple six-class subset configurations demonstrate consistent performance improvements: all classes achieve quantifiable error reduction, with minority classes showing average gains of 12.6% in accuracy and 15.3% in F1-score. Results confirm that the framework significantly enhances robustness against long-tailed class distributions and improves recognition of rare emotion categories. This work establishes a new paradigm for imbalanced affective computing.

0 citationsRead paper

Universal Inverse Distillation for Matching Models with Real-Data Supervision (No GANs)

Sep 26, 2025

Generative models—including diffusion models, flow matching, and related frameworks—suffer from slow inference. Existing knowledge distillation methods are either framework-specific or rely on data-free paradigms; incorporating real data typically necessitates complex adversarial training. Method: We propose the first general-purpose one-step distillation framework, unifying diverse matching-based generative models (e.g., diffusion, flow matching, bridge matching, and stochastic interpolation). Grounded in reverse distillation theory, we introduce a trajectory alignment loss that directly integrates real-data supervision—without GANs or discriminators. Contribution/Results: Our method achieves high-fidelity single-step generation across multiple tasks, significantly accelerating inference while preserving cross-model generalizability and stability. It overcomes dual limitations of conventional distillation: dependence on model-specific architectures and restrictive data strategies.

0 citationsRead paper

Faster Than SVD, Smarter Than SGD: The OPLoRA Alternating Update

Sep 24, 2025

While LoRA fine-tuning reduces parameter count and memory usage, it lags behind full low-rank training (SVDLoRA) in performance. Method: We propose OPLoRA, a memory-efficient optimizer based on Alternating Least Squares (ALS) that decouples LoRA optimization into interpretable subproblems; it approximates truncated SVD accuracy within 1–2 iterations without explicitly forming large matrices. OPLoRA is the first to formulate LoRA training as an alternating update framework, unifying preconditioning strategies and introducing a low-rank-estimation-driven momentum mechanism. Contribution/Results: With only ~3× the parameter overhead of standard LoRA, OPLoRA significantly narrows the performance gap with SVDLoRA. It achieves lower memory consumption and superior generalization across diverse benchmarks—including MNIST, CIFAR-100, and RoBERTa-base—demonstrating both efficiency and effectiveness.

0 citationsRead paper

Recent publications

Latest Papers

On the Equivalence of Optimal Transport Problem and Action Matching with Optimal Vector Fields

Oct 31, 2025

Mapping between sequences of continuous probability distributions arises in generative modeling, yet existing optimal transport (OT) methods suffer from high computational cost due to explicit joint distribution estimation or Wasserstein distance computation. Method: We establish a rigorous equivalence between Action Matching (AM) and OT by formulating distributional transport as learning an optimal vector field for a generative ordinary differential equation (ODE). Under compatible boundary conditions and energy functional design, the AM solution inherently satisfies the Kantorovich duality optimality criterion. Contribution/Results: This equivalence reveals that OT’s optimal coupling is fully characterized by a specific class of integrable vector fields—bypassing explicit joint distribution optimization or Wasserstein metric evaluation. Empirically, AM preserves theoretical optimality while achieving substantial computational efficiency gains. Our work provides a novel theoretical foundation and optimization paradigm for flow-matching-based generative modeling.

0 citationsRead paper

Comparison of Unsupervised Metrics for Evaluating Judicial Decision Extraction

Oct 02, 2025

Legal NLP lacks human-annotated data for evaluating semantic block extraction quality in judicial judgments. Method: We systematically benchmark 16 unsupervised metrics across seven legal content extraction tasks, integrating document-level, semantic-level, structural-level, pseudo-labeling, and law-specific measures—including TF-Coherence, Coverage Ratio, Legal Term Density, and LLM-based scoring—validated via bootstrap correlation analysis, Lin’s concordance correlation coefficient, and mean absolute error (MAE) against expert judgments. Results: TF-Coherence and Coverage Ratio achieve the strongest alignment with expert ratings (r > 0.5, MAE < 0.14), significantly outperforming LLM scoring (r = 0.382), thereby exposing LLMs’ limitations in fine-grained legal assessment. This work introduces the first scalable, unsupervised evaluation framework tailored to legal text, balancing computational efficiency and reliability, and enabling annotation-free, automated quality screening for large-scale legal NLP systems.

0 citationsRead paper

Error correction in multiclass image classification of facial emotion on unbalanced samples

Oct 02, 2025

To address recognition bias toward minority classes in facial emotion multi-class classification caused by class imbalance, this paper proposes a deep network integrating LSTM with spatial attention mechanisms to model emotion features from discriminative facial regions. We introduce a novel six-class subset training strategy and incorporate a seventh class—comprising external rare emotion instances—to enable cross-class error correction. Extensive experiments across multiple six-class subset configurations demonstrate consistent performance improvements: all classes achieve quantifiable error reduction, with minority classes showing average gains of 12.6% in accuracy and 15.3% in F1-score. Results confirm that the framework significantly enhances robustness against long-tailed class distributions and improves recognition of rare emotion categories. This work establishes a new paradigm for imbalanced affective computing.

0 citationsRead paper

Universal Inverse Distillation for Matching Models with Real-Data Supervision (No GANs)

Sep 26, 2025

Generative models—including diffusion models, flow matching, and related frameworks—suffer from slow inference. Existing knowledge distillation methods are either framework-specific or rely on data-free paradigms; incorporating real data typically necessitates complex adversarial training. Method: We propose the first general-purpose one-step distillation framework, unifying diverse matching-based generative models (e.g., diffusion, flow matching, bridge matching, and stochastic interpolation). Grounded in reverse distillation theory, we introduce a trajectory alignment loss that directly integrates real-data supervision—without GANs or discriminators. Contribution/Results: Our method achieves high-fidelity single-step generation across multiple tasks, significantly accelerating inference while preserving cross-model generalizability and stability. It overcomes dual limitations of conventional distillation: dependence on model-specific architectures and restrictive data strategies.

0 citationsRead paper

Faster Than SVD, Smarter Than SGD: The OPLoRA Alternating Update

Sep 24, 2025

While LoRA fine-tuning reduces parameter count and memory usage, it lags behind full low-rank training (SVDLoRA) in performance. Method: We propose OPLoRA, a memory-efficient optimizer based on Alternating Least Squares (ALS) that decouples LoRA optimization into interpretable subproblems; it approximates truncated SVD accuracy within 1–2 iterations without explicitly forming large matrices. OPLoRA is the first to formulate LoRA training as an alternating update framework, unifying preconditioning strategies and introducing a low-rank-estimation-driven momentum mechanism. Contribution/Results: With only ~3× the parameter overhead of standard LoRA, OPLoRA significantly narrows the performance gap with SVDLoRA. It achieves lower memory consumption and superior generalization across diverse benchmarks—including MNIST, CIFAR-100, and RoBERTa-base—demonstrating both efficiency and effectiveness.

0 citationsRead paper