I am an undergraduate in Mathematics and Applied Mathematics at Ningbo University (2024–2028). I previously worked as a research assistant in Prof. Wei Zhang’s group at EAST-Lab, Eastern Institute of Technology. My research focuses on embodied intelligence, multimodal learning and model compression. Previous experience includes video-analysis R&D at Yongjiang Laboratory and remote research at TReNDS Center, USA.
I have co-authored five submitted papers, including three core studies: ACVF and PairSelect as first author, and ReInsVLN as second author and core contributor. I received a National Third Prize in the Tencent Kaiwu Reinforcement Learning Competition.
Outside research, I enjoy strength training, hiking and cycling.
ICRA 2027 · Submitted · Second author · Core contributor
Led early exploration and integration of visual compression, contributed to its implementation, and collaboratively refined memory selection. Collaboratively built the high-level / low-level coordination framework, adapted multiple VLN models, and connected model inference to robot execution. Deployed models on Go2 and G1 and contributed to robot experiments.
On 1,839 R2R Val-Unseen episodes, SR reaches 64.27% / 64.44% (+8.48 / +3.37 pp). Matched-observation tests reduce InternVLA-N1 prefill latency by 21.77% and KV storage by 32.25%.
Independently designed and implemented fusion of MRI, AAL regional volumes and BioClinicalBERT text, using subject-level SigLIP alignment with frozen backbones and cross-backbone ablations.
Evaluated on 1,077 ADNI subjects. With cognitive-score inputs excluded, AD AUC improves by 0.144 on average across three backbones; BrainIAC rises from 0.732 to 0.922.
Independently developed local pairing, norm ranking and group quotas after Transformer block 6. Adapted four detector systems without retraining for compression or budget changes.
At 40% token deletion and batch size 32, encoder speedup is 1.22–1.28×. VideoMAEv2 UCF AUC: 80.63% → 81.02%; TimeSformer XD PR-AUC: 76.69% → 77.37%.
Other co-authored papers: ProSeCIL (fourth author) and TRICE (third author), both submitted to ICASSP 2027.
Built a weakly supervised anomaly-detection and VLM-verification pipeline; QLoRA tuning of Qwen3.5-VL improved ECVA test accuracy from 65.68% to 78.53%.
TReNDS Center · Remote Research Assistant
Georgia State University · Georgia Institute of Technology · Emory University, USA
Conducted ACVF research on multimodal brain MRI, including method design, implementation, cross-backbone experiments and modality ablations.
Tencent Kaiwu · Reinforcement Learning Competition · National Third Prize
Designed a 17-token Transformer policy and PPO training with three value estimates, GAE, BFS escape features and rewards for survival, collection and exploration.
Eastern Institute of Technology, Ningbo (EIT) · Research Assistant
EAST-Lab · Prof. Wei Zhang’s group
Worked on ReInsVLN: led early exploration and integration of visual compression, contributed to its implementation, collaboratively built high-level / low-level coordination, deployed models on Go2/G1, and participated in robot experiments.
Technical skills & methods
I use Python, PyTorch, Transformers and PEFT/QLoRA for model training and adaptation, with experience in SigLIP contrastive learning, PPO, weakly supervised MIL and visual-token compression. My work includes Habitat/VLN-CE evaluation, Unitree SDK2 integration, and brain-imaging pipelines with MONAI and NiBabel. Engineering tools include OpenCV, FastAPI/Flask, SQLite, NumPy/pandas, scikit-learn/XGBoost, Linux/SSH and web interfaces.
