Hello, I'm Che Liu

I am a final-year PhD student at Imperial College London, supervised by Prof. Rossella Arcucci and Prof. Wenjia Bai. I was a visiting student at Technical University of Munich (TUM), advised by Prof. Dr. Daniel Rückert.

I have also conducted research at X-Humanoid on Embodied Understanding and Generation, StepFun on Omni-modal learning, DAMO Academy on Visual Pre-training, and AstraZeneca Cambridge on VLM.

I am actively seeking a full-time research position based in the US, UK, or China, with a focus on multimodal learning, unified visual understanding and generation, and their applications.

Email: che.liu21 [at] imperial.ac.uk

CV Google Scholar LinkedIn

Selected Publications

† - Equal contribution, ‡ - Supervision. More on Google Scholar.

Pelican-VL 1.0: A Foundation Brain Model for Embodied Intelligence

C. Liu, X-Humanoid

Technical Report (X-humanoid), 2025

Nexus-O: An Omni-Perceptive And Interactive Model for Language, Audio, And Vision

C. Liu, et al.

ACM Multimedia 2025

Step-Audio 2

StepFun Audio Team

Technical Report (StepFun), 2025

CogDoc: Towards Unified thinking in Documents

Q. Xu, H. Wang, C. Liu, F. Lin, W. Chen

Technical Report, 2025

Emergent Hierarchical Reasoning in LLMs through Reinforcement Learning

H. Wang, C. Liu, W. Chen, et al.

ICLR, 2026

SRPO: Enhancing Multimodal LLM Reasoning via Reflection-Aware Reinforcement Learning

Z. Wan, C. Liu, S. Yan, et al.

NeurIPS 2025

Does DINOv3 Set a New Medical Vision Standard?

C. Liu, et al.

Technical Report, 2025

NOVA: A Benchmark for Anomaly Localization and Clinical Reasoning in Brain MRI

Cosmin I. Bercea, C. Liu, et al.

NeurIPS Dataset and Benchmark Track 2025 (Oral)

Beyond Distillation: Pushing the Limits of Medical LLM Reasoning with Minimalist Rule-Based RL

C. Liu, et al.

NeurIPS GenAI4Health Workshop 2025

T3D: Advancing 3D Medical Vision-Language Pre-training by Learning Multi-View Visual Consistency

C. Liu, et al.

ICCV VLM3D Workshop 2025 (Oral)

M3Ret: Unleashing Zero-shot Multimodal Medical Image Retrieval via Self-Supervision

C. Liu, et al.

Technical Report (DAMO Academy), 2025

Knowledge-Enhanced Multimodal ECG Representation Learning with Arbitrary-Lead Inputs

C. Liu, et al.

EMNLP Findings 2025

SuPreME: A Supervised Pre-training Framework for Multimodal ECG Representation Learning

M. Cai, J. Jiang, W. Huang, C. Liu‡, R. Arcucci

EMNLP Findings 2025

An Electrocardiogram Foundation Model Built on over 10 Million Recordings with External Evaluation across Multiple Domains

J. Li, A. Aguirre, J. Moura, C. Liu, L. Zhong, C. Sun, G. Clifford, B. Westover, S. Hong

NEJM AI 2024

Can Medical Vision-Language Pre-training Succeed with Purely Synthetic Data?

C. Liu, et al.

ACL Findings 2025

Argus: Benchmarking and Enhancing Vision-Language Models for 3D Radiology Report Generation

C. Liu, et al.

ACL Findings 2025

MEIT: Multimodal Electrocardiogram Instruction Tuning on Large Language Models for Report Generation

Z. Wan†, C. Liu†, et al.

ACL Findings 2025

MedVLM-R1: Incentivizing Medical Reasoning Capability of Vision-Language Models (VLMs) via Reinforcement Learning

J. Pan†, C. Liu†, et al.

MICCAI 2025 (Early Accept)

Zero-Shot ECG Classification with Multimodal Learning and Test-time Clinical Knowledge Enhancement

C. Liu, et al.

ICML 2024

G2D: From Global to Dense Radiography Representation Learning via Vision-Language Pre-training

C. Liu, et al.

NeurIPS 2024

IMITATE: Clinical Prior Guided Hierarchical Vision-Language Pre-training

C. Liu, et al.

IEEE Transactions on Medical Imaging 2024

Med-UniC: Unifying Cross-Lingual Medical Vision-Language Pre-Training by Diminishing Bias

Z. Wan†, C. Liu†, et al.

NeurIPS 2023

M-FLAG: Medical Vision-Language Pre-training with Frozen Language Models and Latent Space Geometry Optimization

C. Liu, et al.

MICCAI 2023

Spectral Cross-Domain Neural Network with Soft-adaptive Threshold Spectral Enhancement

C. Liu, et al.

IEEE TNNLS 2023

Frozen Language Model Helps ECG Zero-Shot Learning

J. Li†, C. Liu†, et al.

MIDL 2023 (Oral)

X-humanoid, Remote Visiting Researcher (Embodied VLM) — Apr 2025 to Present Leading VLM post-training (SFT + self-evolving RL) on large-scale video.
StepFun, Remote Visiting Researcher (OmniLLM) — May 2025 to Present Audio-visual-language SFT/RL and omni-capability benchmarks.
DAMO Academy, Beijing Research Intern — Nov 2024 to Apr 2025 Unified multimodal vision pretraining across 2D/3D/video.
AstraZeneca, Cambridge, UK Research Intern (Vision-Language Models) — Jul 2024 to Sep 2024 Synthetic-data VLP; showed synthetic pretraining can beat real-data baselines.

Imperial College London, UK Ph.D. in Multimodal Learning — Feb 2022 to 2026 (expected) Supervisors: Rossella Arcucci, Wenjia Bai.
Technical University of Munich, Germany Visiting Ph.D. — Apr 2024 to Jun 2024 Supervisor: Daniel Rückert.
Swansea University, UK M.Sc. (Distinction) in Computational Mechanics — Sep 2019 to Sep 2021. Supervisor: Dunhui Xiao.
Shanghai University of Engineering and Science, China B.Sc. in Automotive Engineering — Sep 2016 to Jul 2018.

Hello, I'm Che Liu

Selected Publications

Pelican-VL 1.0: A Foundation Brain Model for Embodied Intelligence

Nexus-O: An Omni-Perceptive And Interactive Model for Language, Audio, And Vision

Step-Audio 2

CogDoc: Towards Unified thinking in Documents

Emergent Hierarchical Reasoning in LLMs through Reinforcement Learning

SRPO: Enhancing Multimodal LLM Reasoning via Reflection-Aware Reinforcement Learning

Does DINOv3 Set a New Medical Vision Standard?

NOVA: A Benchmark for Anomaly Localization and Clinical Reasoning in Brain MRI

Beyond Distillation: Pushing the Limits of Medical LLM Reasoning with Minimalist Rule-Based RL

T3D: Advancing 3D Medical Vision-Language Pre-training by Learning Multi-View Visual Consistency

M3Ret: Unleashing Zero-shot Multimodal Medical Image Retrieval via Self-Supervision

Knowledge-Enhanced Multimodal ECG Representation Learning with Arbitrary-Lead Inputs

SuPreME: A Supervised Pre-training Framework for Multimodal ECG Representation Learning

An Electrocardiogram Foundation Model Built on over 10 Million Recordings with External Evaluation across Multiple Domains

Can Medical Vision-Language Pre-training Succeed with Purely Synthetic Data?

Argus: Benchmarking and Enhancing Vision-Language Models for 3D Radiology Report Generation

MEIT: Multimodal Electrocardiogram Instruction Tuning on Large Language Models for Report Generation

MedVLM-R1: Incentivizing Medical Reasoning Capability of Vision-Language Models (VLMs) via Reinforcement Learning

Zero-Shot ECG Classification with Multimodal Learning and Test-time Clinical Knowledge Enhancement

G2D: From Global to Dense Radiography Representation Learning via Vision-Language Pre-training

IMITATE: Clinical Prior Guided Hierarchical Vision-Language Pre-training

Med-UniC: Unifying Cross-Lingual Medical Vision-Language Pre-Training by Diminishing Bias

M-FLAG: Medical Vision-Language Pre-training with Frozen Language Models and Latent Space Geometry Optimization

Spectral Cross-Domain Neural Network with Soft-adaptive Threshold Spectral Enhancement

Frozen Language Model Helps ECG Zero-Shot Learning

Experiences

Education