Xin Liang

Xin Liang (梁馨)

B.E. in CS @ Tongji University

CV

More details in PDF: CV

Education

Tongji University, Shanghai, China
B.E. in Computer Science, GPA: 90.4/100, IELTS: 7.0
Sep. 2021 – Jul. 2025

Experience

Research Intern, Stony Brook University

  • Sparse adaptation (STAN) with input-dependent routing; consistent gains on LMs/VLMs.
  • SlideGen: multi-agent paper-to-slides pipeline with layout planning & asset matching.

Research Intern, University of Rochester

  • Synthetic data & fairness; subgroup-bias analysis across generations.
  • Self-consuming loop to co-train generator/classifier for robust evaluation.

Research Intern, MIT City Science Lab @ Shanghai

  • Deployed SCRBT token; Solidity contracts on Sepolia with Truffle/Hardhat.
  • Modular smart contracts with ABDK math; end-to-end DApp workflows.

Research Intern, NaMI-Tongji Lab

  • Differentiable NAS via zeroth-order approximation; ICASSP 2023 poster.

Technical Skills

Languages: Python, C/C++, SQL, JavaScript, HTML/CSS
Model Serving & Inference: vLLM (OpenAI-compatible API), FastAPI
Model Training & Fine-tuning: Hugging Face Transformers, ms-swift (ModelScope Swift), PEFT
Frameworks: React, Node.js, Flask, robosuite
Developer Tools: Git, Docker, VS Code, Visual Studio, PyCharm, Linux
Libraries: pandas, NumPy, Matplotlib

Publications

  1. SlideGen: A Multi-Agent Framework for Automatic Scientific Slide Generation — Xin Liang, Xiang Zhang, Yiwei Xu, Siqi Sun, Chenyu You. In submission, 2025.
  2. From Compression to Selection: Sparse Adaptation for Fine-Tuning Large Models — Yifan Wang*, Xin Liang*, Aosong Feng, Haocheng Dai, Tiansheng Wen, Lixuan Guo, Yifei Wang, Chenyu You. In submission, 2025. (*co-first)
  3. Will the Inclusion of Generated Data Amplify Bias Across Generations in Future Image Classification Models? — Zeliang Zhang*, Xin Liang*, Mingqiang Feng, Susan Liang, Chenliang Xu. In submission. (*co-first)
  4. Biomedical Visual Instruction Tuning with Clinician Preference Alignment (BioMed-VITAL) — Hejie Cui*, Lingjun Mao*, Xin Liang, et al. NeurIPS 2024 Datasets & Benchmarks. URL: BioMed-VITAL.
  5. FARPLS: A Feature-Augmented Robot Trajectory Preference Labeling System… — Hanfang Lyu, Xin Liang, et al. IUI 2024. DOI: 10.1145/3640543.3645145.

Honors & Awards

  • Mobile App Innovation Competition — Second Prize, Sep. 2023
  • Asia Pacific Cup Mathematical Modeling — Second Prize, May 2023
  • Shanghai College Student Mathematical Modeling — Third Prize, Nov. 2022
  • “Internet+” Innovation and Entrepreneurship Competition — Bronze Award, Sep. 2022