Selected work

Representative Papers

Target-Guided Selective Reweighting for Physics-Informed Neural Network Inverse Problems: A Transfer Learning Approach

Jul 06, 2026

This work addresses the negative transfer problem in physics-informed neural networks (PINNs) when applied to inverse problems of partial differential equations (PDEs), where discrepancies in physical mechanisms, parameters, or noise levels between source and target domains degrade performance. To mitigate this, the authors propose TGSR-PINN, which reuses only the network weights from a pre-trained source PINN while independently initializing target-domain physical parameters. The method introduces a novel neuron-level scoring mechanism that combines first-order gradient sensitivity with pre-activation variance to assess neuron relevance. Using a Gaussian mixture model, it generates weak adaptation signals to selectively apply soft attenuation to low-scoring neurons. Experiments demonstrate that TGSR-PINN significantly improves the accuracy of target parameter recovery—without compromising field prediction fidelity—in challenging scenarios including high-Péclet-number convection-diffusion, cross-PDE-family transfer from Allen–Cahn to Burgers equations, and reaction-diffusion tasks with 5% noise.

0 citationsRead paper

Seed Hijacking of LLM Sampling and Quantum Random Number Defense

May 08, 2026

This work identifies a critical security vulnerability in the autoregressive sampling process of large language models (LLMs), which relies on deterministic pseudorandom number generators (PRNGs) and is thus susceptible to supply chain attacks. The authors introduce SeedHijack, the first backdoor attack that precisely injects target tokens by manipulating the PRNG seed without altering the model itself. The attack achieves a 99.6% success rate on GPT-2 and 100% success across multiple aligned models ranging from 1.5B to 7B parameters. To counter this threat, the paper proposes a practical defense mechanism leveraging a quantum random number generator (QRNG), which incurs only a 0.6% latency overhead and 7.7 MB of additional memory while effectively neutralizing such attacks.

0 citationsRead paper

DBAW-PIKAN: Dynamic Balance Adaptive Weight Kolmogorov-Arnold Neural Network for Solving Partial Differential Equations

Dec 25, 2025

To address gradient flow stiffness and spectral bias in physics-informed neural networks (PINNs) when solving multiscale or high-frequency partial differential equations (PDEs), this work proposes the Physics-Informed Kolmogorov–Arnold Network (PI-KAN). Methodologically: (i) it introduces a novel KAN architecture integrated with learnable B-spline activations to enhance multiscale representation capacity; (ii) it designs an adaptive loss weighting scheme with dynamically decaying upper bounds to jointly regulate the optimization dynamics of physical constraints and data fidelity; and (iii) it achieves gradient flow stabilization without additional computational overhead. Benchmark evaluations on the Klein–Gordon, Burgers, and Helmholtz equations demonstrate significantly accelerated convergence, over one-order-of-magnitude improvement in accuracy, and markedly enhanced generalization. This work breaks fundamental PINN bottlenecks through dual innovations—network architecture and optimization mechanism.

0 citationsRead paper

Teaching AI Stepwise Diagnostic Reasoning with Report-Guided Chain-of-Thought Learning

Sep 08, 2025

This work addresses the limited clinical reasoning capability and poor interpretability of general-purpose vision-language models (VLMs) in radiological diagnosis. We propose a weakly supervised paradigm that requires no annotated image-lesion pairs, leveraging only free-text radiology reports. Our method automatically parses unstructured reports into structured, stepwise Chain-of-Thought (CoT) reasoning paths, then integrates contrastive image-report alignment with multi-granularity clinical reward-guided reinforcement fine-tuning. To our knowledge, this is the first framework to distill stepwise diagnostic supervision signals—aligned with radiologists’ cognitive reasoning—from raw text reports alone. Zero-shot evaluation on MIMIC-CXR demonstrates substantial improvements: +0.24 in disease classification AUC, +0.23 in lesion localization mIoU, and +0.22 in report generation BLEU score, outperforming state-of-the-art methods. The approach establishes a novel, interpretable, and scalable paradigm for training medical VLMs.

0 citationsRead paper

Coherent Multimodal Reasoning with Iterative Self-Evaluation for Vision-Language Models

Aug 04, 2025

Existing vision-language models (LVLMs) lack deep, chain-of-thought reasoning capabilities for complex cross-modal commonsense inference, often relying on superficial correlations rather than rigorous multi-step deduction. To address this, we propose CoMIR—a coherent multimodal inference framework with iterative self-evaluation. CoMIR employs a three-module architecture: a reasoning decomposition unit, a contextual inference engine, and a coherence assessment module—enabling problem decomposition, progressive inference, and consistency self-checking. Built upon LLaVA-1.6-34B and trained on our newly constructed MDAR dataset, CoMIR supports multi-step reasoning and confidence-aware self-assessment. On benchmarks including VCR, A-OKVQA, and DailyLife-MRC, it achieves state-of-the-art performance among open-source models, attaining an average accuracy of 69.4%—2.4 percentage points higher than the strongest baseline—while significantly improving logical coherence and accuracy in complex multimodal reasoning tasks.

0 citationsRead paper

Recent publications

Latest Papers

Target-Guided Selective Reweighting for Physics-Informed Neural Network Inverse Problems: A Transfer Learning Approach

Jul 06, 2026

This work addresses the negative transfer problem in physics-informed neural networks (PINNs) when applied to inverse problems of partial differential equations (PDEs), where discrepancies in physical mechanisms, parameters, or noise levels between source and target domains degrade performance. To mitigate this, the authors propose TGSR-PINN, which reuses only the network weights from a pre-trained source PINN while independently initializing target-domain physical parameters. The method introduces a novel neuron-level scoring mechanism that combines first-order gradient sensitivity with pre-activation variance to assess neuron relevance. Using a Gaussian mixture model, it generates weak adaptation signals to selectively apply soft attenuation to low-scoring neurons. Experiments demonstrate that TGSR-PINN significantly improves the accuracy of target parameter recovery—without compromising field prediction fidelity—in challenging scenarios including high-Péclet-number convection-diffusion, cross-PDE-family transfer from Allen–Cahn to Burgers equations, and reaction-diffusion tasks with 5% noise.

0 citationsRead paper

Seed Hijacking of LLM Sampling and Quantum Random Number Defense

May 08, 2026

This work identifies a critical security vulnerability in the autoregressive sampling process of large language models (LLMs), which relies on deterministic pseudorandom number generators (PRNGs) and is thus susceptible to supply chain attacks. The authors introduce SeedHijack, the first backdoor attack that precisely injects target tokens by manipulating the PRNG seed without altering the model itself. The attack achieves a 99.6% success rate on GPT-2 and 100% success across multiple aligned models ranging from 1.5B to 7B parameters. To counter this threat, the paper proposes a practical defense mechanism leveraging a quantum random number generator (QRNG), which incurs only a 0.6% latency overhead and 7.7 MB of additional memory while effectively neutralizing such attacks.

0 citationsRead paper

DBAW-PIKAN: Dynamic Balance Adaptive Weight Kolmogorov-Arnold Neural Network for Solving Partial Differential Equations

Dec 25, 2025

To address gradient flow stiffness and spectral bias in physics-informed neural networks (PINNs) when solving multiscale or high-frequency partial differential equations (PDEs), this work proposes the Physics-Informed Kolmogorov–Arnold Network (PI-KAN). Methodologically: (i) it introduces a novel KAN architecture integrated with learnable B-spline activations to enhance multiscale representation capacity; (ii) it designs an adaptive loss weighting scheme with dynamically decaying upper bounds to jointly regulate the optimization dynamics of physical constraints and data fidelity; and (iii) it achieves gradient flow stabilization without additional computational overhead. Benchmark evaluations on the Klein–Gordon, Burgers, and Helmholtz equations demonstrate significantly accelerated convergence, over one-order-of-magnitude improvement in accuracy, and markedly enhanced generalization. This work breaks fundamental PINN bottlenecks through dual innovations—network architecture and optimization mechanism.

0 citationsRead paper

Teaching AI Stepwise Diagnostic Reasoning with Report-Guided Chain-of-Thought Learning

Sep 08, 2025

This work addresses the limited clinical reasoning capability and poor interpretability of general-purpose vision-language models (VLMs) in radiological diagnosis. We propose a weakly supervised paradigm that requires no annotated image-lesion pairs, leveraging only free-text radiology reports. Our method automatically parses unstructured reports into structured, stepwise Chain-of-Thought (CoT) reasoning paths, then integrates contrastive image-report alignment with multi-granularity clinical reward-guided reinforcement fine-tuning. To our knowledge, this is the first framework to distill stepwise diagnostic supervision signals—aligned with radiologists’ cognitive reasoning—from raw text reports alone. Zero-shot evaluation on MIMIC-CXR demonstrates substantial improvements: +0.24 in disease classification AUC, +0.23 in lesion localization mIoU, and +0.22 in report generation BLEU score, outperforming state-of-the-art methods. The approach establishes a novel, interpretable, and scalable paradigm for training medical VLMs.

0 citationsRead paper

Coherent Multimodal Reasoning with Iterative Self-Evaluation for Vision-Language Models

Aug 04, 2025

Existing vision-language models (LVLMs) lack deep, chain-of-thought reasoning capabilities for complex cross-modal commonsense inference, often relying on superficial correlations rather than rigorous multi-step deduction. To address this, we propose CoMIR—a coherent multimodal inference framework with iterative self-evaluation. CoMIR employs a three-module architecture: a reasoning decomposition unit, a contextual inference engine, and a coherence assessment module—enabling problem decomposition, progressive inference, and consistency self-checking. Built upon LLaVA-1.6-34B and trained on our newly constructed MDAR dataset, CoMIR supports multi-step reasoning and confidence-aware self-assessment. On benchmarks including VCR, A-OKVQA, and DailyLife-MRC, it achieves state-of-the-art performance among open-source models, attaining an average accuracy of 69.4%—2.4 percentage points higher than the strongest baseline—while significantly improving logical coherence and accuracy in complex multimodal reasoning tasks.

0 citationsRead paper