CV

Research Interests

Computer Vision, Embodied AI, World Models.
Compositional Generative Model, Scene Understanding, Spatial Reasoning.

Education

ShanghaiTech University — Shanghai, CHN
B.E. in Computer Science and Technology, Minor in Life Sciences | Sep. 2023 – Jun. 2027 (expected)

• AI Honor Class
• Overall GPA: 3.82/4.0 (Rank 9/173 in CS major)
• General Evaluation Ranking: Rank 1/173 in CS major
• Relevant Coursework: Intro. to Information Science and Technology (A+), Intro. to Programming (A), Algorithms and Data Structures (A), Intro. to Machine Learning (A-), AI in Medical Imaging (A), Computational Science and Engineering (A+), Computer Architecture (A-) & Project (A), Protein Design (A+), Game Theory (A)

Harvard University — Cambridge, USA
Visiting Undergraduate Student | Sep. 2025 – May 2026

• Overall GPA: 4.0/4.0
• Relevant Coursework: Signal Processing (MIT cross-registration) (A), Computer Vision (A), Planning and Learning Methods in AI (A), Hardware Architecture for Deep Learning (MIT cross-registration) (in progress), AI for Molecular Biology (in progress), High Performance Computing (in progress)

Awards & Honors

• Merit Student (Top 2% in school), ShanghaiTech University, 2023–2024
• Merit Student (Top 1 in CS major), ShanghaiTech University, 2024–2025
• AI Honor Class (Honors Degrees), ShanghaiTech University, 2024–2027 (expected)
• Gold Medal, International Genetically Engineered Machine Competition (iGEM), 2024
• First Place, Analytical Performance, SensUs Competition, 2025
• Outstanding Mentor Assistant, ShanghaiTech University, 2023

Research Experience

Embodied Minds Lab, Harvard University & Kempner Institute — Cambridge, USA
Visiting Undergraduate Research Assistant | Sep. 2025 – present
Supervisors: Prof. Yilun Du & Dr. Ruojin Cai

• Developed an inverse generative modeling framework for scene understanding on both synthetic and real-world datasets, training a relation-conditioned composable diffusion model that generates scenes from structured object-attribute and spatial-relation inputs.
• Focused on unsupervised object-relation discovery with Diffusion Models, enabling generative models to perform text-image attribution analysis on both synthesized and real-world images.
• Investigated the potential of improving compositional generation with unsupervised relation discovery in a training-free manner, achieving SOTA performance on the 2D spatial relation dataset VISOR.

Perception, Learning and UnderStanding (PLUS) Lab, ShanghaiTech — Shanghai, CHN
Undergraduate Research Assistant | Jan. 2025 – Sep. 2025
Supervisor: Prof. Xuming He

• Investigated Compositional Scene Generation with scene-graph-based diffusion models.
• Focused on learning disentangled representations from scene graphs, enabling flexible control over the generation process.
• Explored Classifier-Free Guidance and training-free methods in image generation.

Selected Competitions

PACIFY – iGEM 2024 [wiki] | Dec. 2023 – Oct. 2024
Team Member

• Performed homology modeling to obtain protein structures and used AlphaFold 2 for structure prediction.
• Operated protein preparation and molecular dynamics simulation.
• Developed devices based on PID algorithm to address itchiness without harming the skin.

MakeSense, ShanghaiTech First SensUs Team | Aug. 2024 – Aug. 2025
Co-Founder & Leader of Data Analysis Team

• Developed a wearable biosensor-based device to continuously monitor acute kidney injury (AKI) biomarkers.
• Invested in an enzyme-based creatinine sensor and QCM (Quartz Crystal Microbalance) platform.
• Set up a data analysis pipeline to process sensor data, achieving near-perfect accuracy in predicting creatinine concentration.
• First Place in Analytical Performance: “This team did an absolutely remarkable job with unprecedented results and near-perfect accuracy. This is a first in SensUs history and sets a new benchmark for other teams, especially as a first-time participating team.” — SensUs Committee.

Selected Projects

De Novo Design of Odorant Binding Proteins for Breast Cancer Detection | Dec. 2024 – Jan. 2025
Course Project, Supervisor: Prof. Jiayi Dou

• Designed three novel Odorant Binding Proteins (OBPs) to specifically recognize VOCs (hexanal, octanal, nonanal) that serve as biomarkers for breast cancer.
• Executed a complete de novo computational design pipeline (RFdiffusionAA + LigandMPNN).
• Validated designs using AutoDock, PyRosetta, and ESMFold, demonstrating significantly higher binding affinity and stability.

Neural Olfactory Sensing and Evaluation (NOSE) | May. 2025 – Jun. 2025
Course Project for Intro. to Machine Learning, Supervisor: Prof. Yujiao Shi

• Fine-tuned MoLFormer, a large chemical language model, on the GS-LF olfactory dataset for specialized odor prediction tasks.
• Evaluated against OpenPOM on Keller-2016, performing odor label classification and pleasantness rating prediction.
• Achieved state-of-the-art performance, matching or surpassing OpenPOM on key metrics.

Phase-Adaptive Quantization for AI Accelerators | Apr. 2026 – May 2026
Course Project for Hardware Architecture for Deep Learning, Supervisor: Prof. Joel Emer & Prof. Vivienne Sze (MIT)

• Designed a phase-aware 4-bit quantization design-space exploration framework for LLM, VLM, and VLA inference workloads, comparing prefill and decode regimes across NVFP4-like, MXFP4-like, and custom rescale pipelines.
• Built an automated AccelForge experiment flow to generate workload/architecture YAMLs, run hardware energy/latency sweeps, extract per-einsum bottleneck breakdowns, and resume long-running Docker-based mappings.
• Implemented a Python quantization accuracy emulator using real model tensor snapshots to evaluate per-layer cosine similarity and combine accuracy with hardware cost for Pareto frontier analysis.
• Demonstrated that decode and prefill favor different quantization configurations, motivating phase-adaptive datapaths over a single fixed 4-bit format.

Technical Strengths

Publications

No publications yet.