Author name cluster

Ye 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.

6 papers

1 author row

AAAI Conference 2026 Conference Paper

RcAE: Recursive Reconstruction Framework for Unsupervised Industrial Anomaly Detection

Rongcheng Wu
Hao Zhu
Shiying Zhang
Mingzhe Wang
Zhidong Li
Hui Li
Jianlong Zhou
Jiangtao Cui

Unsupervised industrial anomaly detection requires accurately identifying defects without labeled data. Traditional autoencoder-based methods often struggle with incomplete anomaly suppression and loss of fine details, as their single-pass decoding fails to effectively handle anomalies with varying severity and scale. We propose a recursive architecture for autoencoder (RcAE), which performs reconstruction iteratively to progressively suppress anomalies while refining normal structures. Unlike traditional single-pass models, this recursive design naturally produces a sequence of reconstructions, progressively exposing suppressed abnormal patterns. To leverage this reconstruction dynamics, we introduce a Cross Recursion Detection (CRD) module that tracks inconsistencies across recursion steps, enhancing detection of both subtle and large-scale anomalies. Additionally, we incorporate a Detail Preservation Network (DPN) to recover high-frequency textures typically lost during reconstruction. Extensive experiments demonstrate that our method significantly outperforms existing non-diffusion methods, and achieves performance on par with recent diffusion models with only 10% of their parameters and offering substantially faster inference. These results highlight the practicality and efficiency of our approach for real-world applications.

PDF Details DOI

EAAI Journal 2025 Journal Article

A cross-dimensional synergistic network for brain tumor segmentation

Chih-Wei Lin
Ye Lin

Brain tumor segmentation is crucial in medical image processing but struggles to access and integrate information across dimensions, such as global–local, multi-scale, long-range dependencies information, and the receptive field, leading to suboptimal accuracy and robustness. This study proposes a novel attention network, the Cross-Dimensional Synergistic Network (CS-Net), for brain tumor segmentation, which incorporates three kinds of symmetrical relationships — also called cross-dimensional information, including global–local, vertical-horizontal axis, and large–small (multi-scale) relationships. The symmetrical relationships are translated into three attention mechanisms: Global–local Region Attention (GRA), Axis-aligned Translation Attention (ATA), and Multi-scale Fusion Attention (MFA). The GRA module enhances spatial perception by dividing the input feature map into non-overlapping sub-regions and computing attention weights between each sub-region (Local) and global features to capture finer spatial dependencies. The ATA module calculates the cross-directional attention between the entire original and shifted features, where the shifted features include horizontal and vertical shifts to learn the long-range dependencies and extend the receptive field in different directions. The MFA module executes self-attention interactions between feature maps at different scales to effectively integrate adjacent scale information. Experiments on the Brain Tumor Segmentation Challenge (BraTS) from the Medical Image Computing and Computer-Assisted Intervention (MICCAI) and the Medical Segmentation Decathlon (MSD) demonstrate significant performance improvements. Our model achieves 86. 23%, 89. 99%, and 1. 2485 in the Dice coefficient, precision, and 95%Hausdorff Distance on the MICCAI BraTS dataset and achieves 81. 57%, 86. 70%, and 1. 5547 in the Dice coefficient, precision, and 95%Hausdorff Distance on the MSD BraTS dataset, suppressing the state-of-the-art approaches.

Details DOI

EAAI Journal 2025 Journal Article

A lightweight wood defect segmentation network via multi-dimension boundary perception and guidance

Yuhang Zhu
Ye Lin
Zhezhuang Xu
Dan Chen
Kunxin Zheng
Yazhou Yuan

Wood surface defect segmentation is extremely critical for defect refinement and quality control of wooden products. However, it is a challenging task to develop an efficient method with current algorithms due to the complicated characteristics of wood defects with obscure boundary, intraclass difference and interclass similarity. To address these issues, a lightweight network via multi-dimension boundary perception and guidance is proposed for precise segmentation of wood defects. At first, based on the Segformer, a boundary prediction branch is added to enrich detailed boundary information in the encoder, and supervised by the Gaussian signal and cosine similarity, to balance the effect of the boundary gradient information. Then, a double-flow enhancing module is designed to integrate the adjacent level features, by embedding two enhancing paths, to adaptively generate discriminative information of the defects. Finally, a binary segmentation head following the predicted map is introduced to strengthen the penalty for the false prediction results of the boundary. Experimental results demonstrate the proposed method outperforms the state-of-the-arts on our wood surface defect dataset, as well as on three public datasets.

Details DOI

AIIM Journal 2025 Journal Article

Toward fair medical advice: Addressing and mitigating bias in large language model-based healthcare applications

Haohui Lu
Ye Lin
Zhidong Li
Man Lung Yiu
Yu Gao
Shahadat Uddin

Large Language Models (LLMs) are increasingly deployed in web-based medical advice applications, offering scalable and accessible healthcare solutions. However, their outputs often reflect demographic biases, raising concerns about fairness and equity for vulnerable populations. In this work, we propose FairMed, a framework designed to mitigate biases in LLM-generated medical advice through fine-tuning and prompt engineering strategies. We evaluate FairMed using language-based and content-level metrics across demographic groups on publicly available (MedQA), synthetic (Synthea), and private (CBHS) datasets. Experimental results demonstrate consistent improvements over Llama3 - Med42, as well as over the zero-shot prompting baseline. For instance, in sentiment analysis for gender groups using MedQA, FairMed with Descriptive Prompting reduces the Statistical Parity Difference (SPD) from 0. 0902 to 0. 0658, improves the Disparate Impact Ratio from 1. 1916 to 1. 1566, and decreases the Kullback–Leibler Divergence from 0. 0045 to 0. 0024. Similarly, in directive language evaluation for gender groups using Synthea, SPD improves from 0. 1056 to nearly zero, achieving near-perfect parity. On the CBHS dataset, FairMed with Descriptive Prompting increases Diagnostic Recommendation Divergence (DRD) for race groups from 0. 9530 to 0. 9848, indicating improved group-specific tailoring, while reducing the Action Disparity Index (ADI) from 0. 0857 to 0. 0469 and Referral Frequency Parity (RFP) from 0. 0791 to 0. 0511, reflecting enhanced fairness. These findings highlight FairMed’s effectiveness in addressing demographic disparities and promoting equitable healthcare guidance through web technologies. This framework contributes to building trustworthy and inclusive systems for delivering medical advice by ensuring fairness in sensitive applications.

Details DOI

AAAI Conference 2021 Conference Paper

An Efficient Transformer Decoder with Compressed Sub-layers

Yanyang Li
Ye Lin
Tong Xiao
Jingbo Zhu

The large attention-based encoder-decoder network (Transformer) has become prevailing recently due to its effectiveness. But the high computation complexity of its decoder raises the inefficiency issue. By examining the mathematic formulation of the decoder, we show that under some mild conditions, the architecture could be simplified by compressing its sub-layers, the basic building block of Transformer, and achieves a higher parallelism. We thereby propose Compressed Attention Network, whose decoder layer consists of only one sub-layer instead of three. Extensive experiments on 14 WMT machine translation tasks show that our model is 1. 42× faster with performance on par with a strong baseline. This strong baseline is already 2× faster than the widely used standard baseline without loss in performance.

PDF Details

IJCAI Conference 2020 Conference Paper

Towards Fully 8-bit Integer Inference for the Transformer Model

Ye Lin
Yanyang Li
Tengbo Liu
Tong Xiao
Tongran Liu
Jingbo Zhu

8-bit integer inference, as a promising direction in reducing both the latency and storage of deep neural networks, has made great progress recently. On the other hand, previous systems still rely on 32-bit floating point for certain functions in complex models (e. g. , Softmax in Transformer), and make heavy use of quantization and de-quantization. In this work, we show that after a principled modification on the Transformer architecture, dubbed Integer Transformer, an (almost) fully 8-bit integer inference algorithm Scale Propagation could be derived. De-quantization is adopted when necessary, which makes the network more efficient. Our experiments on WMT16 En Ro, WMT14 En De and En->Fr translation tasks as well as the WikiText-103 language modelling task show that the fully 8-bit Transformer system achieves comparable performance with the floating point baseline but requires nearly 4x less memory footprint.

PDF Details DOI