Ze Rong 戎泽

Ze Rong 戎泽

Stay curious. Build what matters.

I’m Ze Rong (戎泽) — an undergraduate in Computer Science & Technology at Nantong University.
My work sits at the intersection of medical imaging, multimodal learning, and sports analytics. I’m particularly interested in:

  • Frequency-domain representation learning for robust medical image understanding
  • Vision–language alignment and evidence-grounded reasoning for clinical AI
  • Graph-based tactics modeling for soccer using tracking, audio, and commentary
  • Non-invasive brain decoding (fMRI/MEG → language/semantics)
  • Federated learning & unlearning with efficient, economics-inspired training

I enjoy building full pipelines—from data engineering to models to deployment—and writing clean, reproducible research code.

News

  • 2025 — FaRMamba accepted to ICONIP 2025 (medical image segmentation with frequency learning + reconstruction-aided Mamba).
  • Ongoing — EchoGNN for emotion-aware tactical analysis on soccer broadcast + tracking.
  • Ongoing — SMN4Lang-based semantic decoding reproduction, with extensions to frequency-aligned modules.

Selected Projects

FaRMamba — Frequency-based & Reconstruction-aided Mamba for Medical Segmentation

  • Frequency-domain modules + auxiliary reconstruction to improve Dice/robustness.
  • Status: accepted to ICONIP 2025.
  • Code/Preprint: [link]

EchoGNN (sports analytics) — Emotion-aware spatio-temporal graph for counter-attack modeling

  • Fuse player/ball tracks, audio arousal (CLAP/whisper), and commentary semantics; ST-GNN with FiLM.
  • Status: in progress.
  • Code: [link]

FIRM (VI-ReID) — Fusion-Injected Residual Memory for cross-modal person re-identification

  • Token-level alignment + hierarchical fidelity modeling.
  • Status: manuscript.
  • Preprint: [link]

BCI Semantic Decoding — Reproducing & extending HuthLab pipeline

  • ROI-aware features, ridge baselines, frequency-aligned enhancements.
  • Status: in progress.
  • Notes/Code: [link]

Publications & Manuscripts

  • FaRMamba — ICONIP 2025. Frequency-based learning & reconstruction-aided Mamba for medical segmentation.
    paper · code
  • MSC-LSAM — Journal of Data Acquisition and Processing, 2025 (planned/accepted as per your records).
    paper
  • More works in preparation on multimodal medical AI, sports analytics, and BCI decoding.

A compact, up-to-date list lives on [Google Scholar] and [GitHub].

Experience (highlights)

  • Nantong University — Undergraduate researcher, AI & medical imaging lab (with Prof. Lei Ma).
  • Collaborations with football analytics groups on SkillCorner OpenData.
  • Hands-on federated learning & unlearning pipelines; reproducible experiments on 3090 GPUs.

Open-source & Engineering

  • Clean, documented code; reproducible configs; data converters (kloppy→polars), evaluation scripts, and figure generation.
  • Tooling: PyTorch · TensorFlow (Spektral) · OpenMMLab · Whisper/CLAP · Polars · JAX (occasional) · Docker.

Awards (selected)

  • China Robot and AI Competition (CRAIC) 2025 — National Second Prize
  • CRAIC 2024 — National First Prize

Looking ahead

I’m seeking opportunities to pursue a PhD in medical AI / multimodal reasoning, with strong emphasis on frequency-domain learning, VLM alignment, and clinically reliable models. I value teams that care about reproducibility, transparency, and real clinical impact.

If our interests align, let’s talk.