Author name cluster

Qiang Lin

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

5 papers

2 author rows

AAAI Conference 2026 Conference Paper

AP2O-Coder: Adaptively Progressive Preference Optimization for Reducing Compilation and Runtime Errors in LLM-Generated Code

Jianqing Zhang
Wei Xia
Hande Dong
Qiang Lin
Jian Cao

LLM's code generation capabilities have yielded substantial improvements in the effectiveness of programming tasks. However, LLM-generated code still suffers from compilation and runtime errors. Existing offline preference optimization methods primarily focus on enhancing LLMs' coding abilities using pass/fail signals in the preference data, overlooking the deep-level error types in the failed codes. To address this, we propose Adaptively Progressive Preference Optimization (AP2O) for coding (i.e., AP2O-Coder), a method that guides LLMs adaptively and methodically to reduce code errors for code generation. Specifically, we construct an error notebook from failed codes and progressively optimize the LLM to correct errors type by type. Furthermore, we adaptively replay error types to tailor to the LLM's evolving weaknesses throughout training. Through extensive experiments on both code and general LLMs (Llama, Qwen, and DeepSeek series) with parameters ranging from 0.5B to 34B, our AP2O-Coder improves code generation performance by up to 3% in pass@k while using less preference data.

PDF Details DOI

EAAI Journal 2026 Journal Article

Fusion of deep and manual features for improved generation of large-size, low-resolution functional medical images

An Xie
Qiang Lin
Xianwu Zeng
Yongchun Cao
Zhengxing Man
Caihong Liu
Zhengqi Cai
Xiaodi Huang

Deep learning is effective for automated medical image analysis, but it relies heavily on large and diverse datasets. Such datasets are particularly scarce for single-photon emission computed tomography (SPECT) bone scintigrams, which hinders model development. Current generative models often produce samples with limited structural detail and insufficient diversity, reducing their effectiveness for downstream diagnostic tasks. To overcome these limitations, we introduce a high-frequency information fusion generator (HFIF-Gen), a novel SPECT bone scintigram generation model built on a generative adversarial network framework. HFIF-Gen produces samples that are both structurally faithful and feature-diverse, preserving anatomical integrity while enhancing variability. Specifically, our method integrates high-frequency information with an attention mechanism to improve the quality of generated anatomical structures, such as bones. To further improve sample diversity, we employ a two-stage training strategy combined with a Fusion Intensity Factor (FIF), which effectively incorporates multi-view features into the generated samples to better support practical diagnostic applications. Experimental evaluation using 276 real clinical patient SPECT images demonstrates superior performance, with a Peak Signal-to-Noise Ratio (PSNR) of 38. 75 ± 0. 97 (a gain of +2. 38 over the MedGAN) and a Structural Similarity Index Measure (SSIM) of 0. 916 ± 0. 01 (an improvement of 6. 28%). We employed multiple downstream models for further validation. Incorporating these generated samples into diagnostic tasks yields consistent improvements: 3. 04%-17. 24% in Accuracy, 0. 73%-21. 02% in Precision, 7. 14%-12. 14% in Recall, and 4. 50%-15. 99% in F-1 score. These results validate that HFIF-Gen improves both the quality and diversity of generated samples, improving the accuracy of downstream diagnostic tasks.

Details DOI

EAAI Journal 2025 Journal Article

A three-stage segmentation framework for lung cancer lesion isolation in three-dimensional positron emission tomography images

Yusheng Wu
Qiang Lin
Jingjun Wei
Yongchun Cao
Zhengxing Man
Xiaodi Huang

Background Positron emission tomography (PET) is a critical functional medical imaging modality for the early detection and diagnosis of cancers. PET imaging faces several challenges that hinder accurate interpretation including its inherently low spatial resolution, substantial variability in cancer lesions’ appearance, and difficulties distinguishing between the image background and benign lesions. Methods We propose a novel three-stage image segmentation framework to enhance the accuracy of lung cancer lesion identification and extraction from three-dimensional (3D) PET images. The first stage conducts a coarse segmentation using an encoder-decoder structure network to roughly position lesions. The second stage employs a multi-layer feature extraction network to learn the detailed characteristics of coarse segmentation results, mitigating false positives caused by localization inaccuracy. The last stage further refines the extracted features via dividing a sub-region of the lesion into foreground and background branches, reducing false positives caused by over-segmentation of edges. A novel lesion count loss function is introduced to guide the model to generate predictions during the training, ensuring that the predicted lesion counts align with the ground truth labels. Results The proposed method was evaluated on clinical 3D PET image datasets. Experimental results demonstrated a Dice Similarity Coefficient (DSC) of 85. 35 %, Accuracy of 83. 97 %, and Recall of 86. 83 %. Compared to existing models applied to the same datasets, our method consistently achieved superior performance. Conclusion The proposed method significantly improves the segmentation performance of lung cancer lesions, implying that our method holds substantial potential for broader clinical application, even in low-resolution images.

Details DOI

IROS Conference 2025 Conference Paper

A Variable-stiffness Neck Exoskeleton with Pneumatic-driven Actuators for Prolonged Head Flexion Assistance

Tianfang Li
Baojun Chen
Zhichao Hua 0003
Qiang Lin

Surgeons are frequently subjected to prolonged unnatural postures, such as sustained neck flexion during surgical procedures, increasing their susceptibility to work-related musculoskeletal disorders (WMSDs). Despite the prevalence of such issues, there remains a scarcity of ergonomic solutions that effectively mitigate neck pain while permitting full operational functionality across diverse environments. In this paper, we present a novel variable-stiffness neck exoskeleton with pneumatic-driven tensile actuators for prolonged head flexion assistance. Its innovative design facilitates unrestricted head movement while providing expected neck support, thereby minimizing strain on antagonist muscle groups. Comprehensive experimental evaluations were conducted to assess the system’s performance. Transitional response times between flexible and rigid states were 0. 28 s and 0. 46 s, respectively. Experimental trials involving five healthy subjects demonstrated that the average muscular activity reduction of the splenius capitis and the sternocleidomastoid muscles were 38. 8 ± 2. 0% and 9. 7 ± 2. 5%, respectively. These experimental results demonstrated great potential of the exoskeleton in practical application for alleviating the physical burden during prolonged head flexion.

Details

EAAI Journal 2022 Journal Article

Block-based multi-view classification via view-based L 2, p sparse representation and adaptive view fusion

Zhi Wang
Qiang Lin
Yingyi Chen
Ping Zhong

To date, many efficient classification methods have been presented by utilizing the multi-view data’s rich information. Nevertheless, they commonly construct models by concatenating entire views together into the high-dimensional vectors while ignoring the individuality and relationship of views. Also, they often use fixed labels to perform classification, ignoring the requirement of the large margin between distinct classes. To address the above problems, we propose a new block-based multi-view classification model via view-based L 2, p sparse representation and adaptive view fusion. Specifically, the model establishes the L 2, p regularization in each view space to excavate the individuality information of views. Meanwhile, a newly proposed shared loss term across views is combined in the model to learn the complementarity and consistency information of views. The adaptive weighting is introduced to measure the contribution of distinct views while performing adaptive view fusion. The model also adopts slack labels to increase the distance of distinct classes. Furthermore, an Alternating Direction Method of Multipliers (ADMM) based algorithm is designed to solve the model through block calculation rapidly. And a strict theoretical proof of its convergence is provided. Extensive experiments demonstrate that the proposed method achieves superior performance.

Details DOI