Academic CV

Li Lin 林立

Senior AI Engineer · CATL Hong Kong Research Institute (HKRI)

Research Experience

CATL Hong Kong Research Institute (HKRI)

Nov 2025 – Present

Senior AI Engineer

Recursive Self-Improvement (RSI) and Automated Agent Harness Optimization

I study how execution traces can inform automated harness revisions. I use controlled evaluations across successive revisions to measure cross-task transfer, regression, and whether improvements are sustained across iterations.

Multi-Agent AutoResearch Frameworks

I develop multi-agent frameworks that coordinate hypothesis development, experimental design, and independent verification for materials research. Applications include machine-learned force fields, materials property prediction, and screening.

Multimodal, generalizable AI for battery characterization

I develop advanced signal processing and multimodal AI methods for ultrasonic measurements, scanning electron microscopy (SEM), and other characterization modalities. I study how representations of battery state and microstructure generalize across cells, instruments, and measurement conditions.

The University of Hong Kong

Sep 2021 – Sep 2025

Doctoral Researcher, OMEGA Lab · Joint PhD training with SUSTech MedICAL Lab

My research in medical imaging and computer vision spans my doctorate and subsequent collaborations.

Data- and annotation-efficient learning

I develop methods that learn from scarce, incomplete, or noisy supervision and imbalanced data. This work combines anatomy-guided synthesis, prototype-based pseudo-label calibration, and semi-supervised adaptation to make better use of unlabeled scans and underrepresented cases.

Cross-domain generalization and adaptation

I study how imaging models transfer across acquisition domains and annotation protocols. My work includes learned augmentation for unseen domains, complementary teachers for adaptation to multiple unlabeled targets, and anatomy-aware prompting for supervision at different levels of detail.

Personalized federated learning across data silos

I developed personalized federated learning methods for institutions with heterogeneous imaging distributions and weak annotations. Site-aware representation learning, shared and local prompts, and adaptive aggregation or distillation let institutions share knowledge while retaining models suited to their own data.

Multimodal representation learning and foundation models

I study transferable representations through masked modeling of 2D and 3D scans, complementary MRI pretraining, and image–report alignment. Related work addresses noisy image–text correspondence and grounds medical vision–language reasoning in clinical knowledge graphs.

Image synthesis and reconstruction

I work on recovering or synthesizing complementary information when acquisitions differ in contrast, resolution, or spatial alignment. Applications include cross-contrast MRI reconstruction, diffusion synthesis for unaligned multiphase CT, and anatomy-guided CT synthesis for PET attenuation correction.

Clinical image analysis and decision support

I contribute to methods that turn imaging evidence into quantitative findings and patient-level assessments. This work spans anatomy-aware measurements, uncertainty-aware PET/CT fusion, 3D OCT diagnosis, and the evaluation of multidisciplinary agents for treatment planning.

3D computer vision

I also contribute to family-aware models for reconstructing animal pose and shape from single images. This work combines shared representations with family-level supervision to transfer across species.

Southern University of Science and Technology

Jul 2019 – Aug 2021

Visiting Student and Research Assistant, MedICAL Lab

I studied lesion detection from coarse annotations, anatomical segmentation, and contrastive image synthesis and enhancement. I was the first-listed author of the SUSTech–SYSU dataset paper in Scientific Data. BSDA-Net first appeared as a preprint during this appointment and was subsequently published at MICCAI 2021.

Education

Sun Yat-sen University

Sep 2018 – Jun 2021

MS in Information and Communication Engineering

Outstanding Thesis nomination (top 5%)

South China Normal University

Sep 2014 – Jun 2018

BEng in Communication Engineering

Outstanding Thesis

13 publications
  1. Full end-to-end diagnostic workflow automation of 3D OCT via foundation model-driven AI for retinal diseasesFull end-to-end diagnostic workflow automation of 3D OCT via foundation model-driven AI for retinal diseasesJinze Zhang*, Jian Zhong*, Li Lin*, Jiaxiong Li*, Ke Ma, Naiyang Li, Meng Li, Yuan Pan, Zeyu Meng, Mengyun Zhou, Shang Huang, Shilong Yu, Zhengyu Duan, Sutong Li, Honghui Xia, Juping Liu, Dan Liang, Yantao Wei, Xiaoying Tang†, Jin Yuan†, Peng Xiao† Co-first author · npj Digital Medicine (JIF 18.0)
  2. An agentic multidisciplinary framework for diabetic retinopathy treatment planningAn agentic multidisciplinary framework for diabetic retinopathy treatment planningTingting Shen, Zhiyuan Cai, Pujin Cheng, Li Lin, Meng Tian, Honghui Xia, Zhihao Jiang, Ting Zhang, Yantao Wei, Zhaoyang Wang, Jin Yuan, Wenjun Xu, Xiaoying Tang · Research Square Preprint
  3. SynLiTS: Phase Prompting-Driven Diffusion Synthesis and Context-Aware Fusion for Unaligned Liver Tumor SegmentationSynLiTS: Phase Prompting-Driven Diffusion Synthesis and Context-Aware Fusion for Unaligned Liver Tumor SegmentationShiyun Chen*, Li Lin*, Zhicheng Jin*, Pujin Cheng, Jianjian Chen, Haidong Zhu, Kenneth K. Y. Wong, Xiaoying Tang† Co-first author · IEEE Transactions on Medical Imaging (JIF 12.4)
  4. Diff4MMLiTS: Advanced Multimodal Liver Tumor Segmentation via Diffusion-Based Image Synthesis and AlignmentDiff4MMLiTS: Advanced Multimodal Liver Tumor Segmentation via Diffusion-Based Image Synthesis and AlignmentShiyun Chen*, Li Lin*, Pujin Cheng, Zhicheng Jin, Jianjian Chen, Haidong Zhu, Kenneth K. Y. Wong, Xiaoying Tang Co-first author · MLMI 2025 at MICCAI
9 publications
  1. Multi-modal evidential fusion network for trustworthy PET/CT tumor segmentationMulti-modal evidential fusion network for trustworthy PET/CT tumor segmentationYuxuan Qi*, Li Lin*†, Bin Zhang, Jingya Zhang, Jiajun Wang Co-first author Co-corresponding author · Knowledge-Based Systems (JIF 8.0)
  2. Test-Time Fluorescein Angiography-Free Non-Perfusion Area Segmentation from Fundus ImageTest-Time Fluorescein Angiography-Free Non-Perfusion Area Segmentation from Fundus ImageJuntao Chen*, Li Lin*, Jiaming Qiu*, Huiyu Liang, Terry Tao Ye, Haoyu Chen†, Xiaoying Tang† Co-first author · IEEE ISBI
  3. LF-SynthSeg: label-free brain tissue-assisted tumor synthesis and segmentationLF-SynthSeg: label-free brain tissue-assisted tumor synthesis and segmentationPengxiao Xu, Junyan Lyu, Li Lin, Pujin Cheng, Xiaoying Tang · IEEE Journal of Biomedical and Health Informatics (JIF 7.7)
14 publications
  1. pFedWSD: unified weakly supervised personalized federated image segmentation via similarity-aware distillationpFedWSD: unified weakly supervised personalized federated image segmentation via similarity-aware distillationJianshan Pan, Li Lin, Jiewei Wu, Yixiang Liu, Xiaohua Chen, Qiyou Lin, Jianye Huang, Xiaoying Tang† First student author · Journal of Image and Graphics (CCF B / T2)
10 publications
  1. GAMMA challenge: Glaucoma grAding from Multi-Modality imAgesGAMMA challenge: Glaucoma grAding from Multi-Modality imAgesJunde Wu, Huihui Fang, Fei Li, Huazhu Fu, et al. (including Li Lin) · Medical Image Analysis (JIF 14.0)
9 publications
10 publications
  1. Fargo: A joint framework for faz and rv segmentation from octa imagesFargo: A joint framework for faz and rv segmentation from octa imagesLinkai Peng*, Li Lin*, Pujin Cheng, Zhonghua Wang, Xiaoying Tang Co-first author · OMIA at MICCAI
3 publications
  1. Automated Diabetic Retinopathy Identification via Lesion Guided NetworkAutomated Diabetic Retinopathy Identification via Lesion Guided NetworkYijin Huang, Li Lin, Meng Li, Jin Yuan, Xiaoying Tang† · International Symposium on Image Computing and Digital Medicine (ISICDM)
  2. The SUSTech-SYSU dataset for automated exudate detection and diabetic retinopathy gradingThe SUSTech-SYSU dataset for automated exudate detection and diabetic retinopathy gradingLi Lin*, Meng Li*, Yijin Huang*, Pujin Cheng, Honghui Xia, Kai Wang, Jin Yuan†, Xiaoying Tang† First author · Scientific Data (JIF 7.2)
1 publications

Patents

Granted Inventions

  • CN118134946B
    Medical image segmentation with learnable prompts and personalized federated aggregationSUSTech
  • CN116229070B
    Federated image segmentation with cyclic knowledge distillationSUSTech
  • CN116740075B
    Personalized federated segmentation with adaptive head aggregationSUSTech
  • CN115965631B
    Weakly supervised curvilinear segmentation with synthetic training imagesSUSTech
  • CN115880250B
    Medical image quality assessment with learned representations and clusteringSUSTech
  • CN110786824B
    Lesion detection from coarse annotations via bounding-box refinementSun Yat-sen University

Published Applications

  • CN121119113A
    Efficient text reasoning with cascaded teacher–student modelsSUSTech
  • CN120689607A
    Multimodal liver tumor segmentation with modality-conditioned synthesisSUSTech
  • CN120689606A
    Multimodal MRI segmentation with complementary masked modelingSUSTech
  • CN120635108A
    Prompt-based segmentation across retinal annotation granularitiesSUSTech
  • CN119540608A
    Image recognition with parameter-efficient adaptation of pretrained modelsSUSTech
  • CN117373102A
    Ophthalmic image analysis with multimodal pretrainingSUSTech
  • CN116206110A
    Fundus image segmentationSUSTech

Open Research

Datasets and Benchmarks

  • HROCT-Bench / UniOCTSeg++

    IEEE TMI 2026

    A retinal OCT layer segmentation benchmark spanning 11 datasets and eight annotation granularities. The release includes six preprocessed subsets and split files. The repository provides data access instructions and supervised and semi-supervised training scripts.

  • YoloCurvSeg Benchmark Data

    MedIA 2023

    Five prepared benchmarks for weakly supervised vessel and nerve segmentation, with dense masks, noisy skeleton labels, and documented data splits. The release supports comparisons across annotation budgets and imaging domains.

  • SUSTech–SYSU

    Scientific Data 2020

    1,219 fundus photographs with exudate annotations, optic disc boxes, fovea locations, and diabetic retinopathy grades from three protocols. The shared annotations support lesion detection and comparisons across grading conventions.

Research Implementations and Models

  • EyeGraph

    IEEE TMI 2026

    Vision–language pretraining and instruction-tuning code for retinal image interpretation, with a knowledge-graph reasoning module for report generation and a recorded demonstration.

  • MF-MCM

    Information Fusion 2026

    Complementary masked pretraining for learning from multiple MRI sequences. The implementation includes pretraining, fine-tuning, and 3D evaluation for brain tumor segmentation.

  • DRAgent

    Research Square 2026 · Preprint

    A multidisciplinary agent implementation for studying treatment planning in diabetic retinopathy. The repository includes specialist agents, an evaluation split, and quantitative scoring scripts.

  • MIRAGE

    MedIA 2026

    Medical image–text pretraining under noisy pairings. The repository provides data preparation, training configurations, and objectives for reducing the effect of mismatched image–report pairs.

  • AniMer+ / AniMer

    IEEE TPAMI 2026 · CVPR 2025

    Family-aware reconstruction of animal pose and shape from a single image. The project page shows predictions for mammals and birds, and both versions provide training and evaluation code.

  • ProCNS

    IEEE JBHI 2025

    Progressive prototype calibration and noise suppression for segmentation from sparse annotations. The release provides processed data links and training and evaluation commands.

  • FedLPPA / FedICRA

    IEEE TMI 2025 · MLMI at MICCAI 2023

    Personalized federated segmentation across sites with different imaging distributions and weak labels. Both implementations document data preparation, client settings, and training and evaluation.

  • PRIOR

    ICCV 2023

    Prototype-based local and global alignment of medical images and reports. The repository includes the pretraining implementation and a released ResNet-50 image encoder.

  • AADG

    IEEE TMI 2022

    Automatic augmentation search for domain generalization in retinal image segmentation. The repository includes policy search, training, and evaluation code for studying transfer to unseen acquisition domains.

Research Tools

  • UU Remote Ubuntu Plus

    Ubuntu GNOME · MIT

    I maintain a NetEase UU bridge for remote development, experiment monitoring, and visualization on Ubuntu. It adds canvas presets up to 4K, Chinese input and clipboard compatibility, and an option to retain terminal sessions across reconnects.

Recognition

Open figure source ↗