Shanghai Jiao Tong University (SJTU), China

M.Sc. in Applied Statistics | Sep. 2021 - Jun. 2023

Southern University of Science and Technology (SUSTech) (Top 10%), China

B.Sc. in Mathematics and Applied Mathematics | Sep. 2017 - Jun. 2021

[Jun. 2023 - Current] Integrated Framework for Multi-objective Task Allocation and Path Finding in Large-scale Robot Network

Advised by Prof. Hesheng Wang and Prof. Zhe Liu

• An integrated task assignment, path planning, and coordination problem was applied for large-scale robot networks with uncertainties, reaching multiple objectives.
• Simultaneously allocate tasks to distinct robots based on multiple objectives: a novel generalized conflict graph, the electricity of robots, the priority of tasks, and the suitability between robots and tasks, etc.
• The enhanced SMTA helps reach higher optimality, scalability, robustness, and efficiency of the integrated optimization problem.

[Feb. 2023 - Jul. 2023] Selective Multi-Robotic Task Allocation (SMTA) [Paper]

Advised by Prof. Hesheng Wang and Prof. Zhe Liu

• Pose greater attention to tasks with significant influence on the entire system and formulate the problem of critical subset identification mathematically and statistically.
• Propose the SMTA algorithm based on the conditional expectation: precisely assigning tasks within the critical task subset while distributing the remaining jobs with simple strategies.
• Experiments demonstrate that SMTA considerably reduces computational requirements in theory, maximizes task allocation efficiency, and shows high extendability.

[Sep. 2022 - Jun. 2023] Traffic Flow Learning Enhanced Large-Scale Multi-Robot Cooperative Path Planning Under Uncertainties

Advised by Prof. Hesheng Wang and Prof. Zhe Liu

• Consider traffic flow equilibrium to reduce congestion and increase efficiency. We propose a prediction-based hierarchic path planning framework, including traffic flow prediction, sector-level planning, and road-level coordination.
• In traffic flow prediction, a spatio-temporal GNN predicts the evolution of robot density distribution. Sector-level paths are generated considering travel distance and traffic flow equilibrium simultaneously. CBS algorithm within each sector ensures conflict-free local paths in road-level.
• By explicitly considering motion/communication uncertainties, our approach well-balances the system-level performance, local motion safety, and real-time performance.

[Sep. 2022 – Apr. 2023] Graph Representation Learning for Interactive Biomolecule Systems [Paper]

Advised by Prof. Yuguang Wang

• Present a comprehensive review of the methodologies used to represent biological molecules and systems as computer-recognizable objects, such as sequences, graphs, and surfaces.
• Examine how geometric deep learning models, especially graph-based techniques, can capture the geometric information of biomolecules at different levels to enable drug discovery, protein characterization, and biological system analysis.
• Conclude with an overview of the current state of biomolecule systems and their mechanisms, highlighting the existing challenges and potential future research directions.

[Jun. 2022 - Nov. 2022] Lightweight Graph Neural Networks in Protein Mutant Design [Paper]

Advised by Prof. Yuguang Wang

• Design a lightweight zero-shot GNN model for multi-task protein representation learning from its 3D structure, tackling the problem of directed evolution in computational protein design.
• Creatively train the model through denoising: recovering the amino acid types and crucial properties of the central residues from a given noisy 3D local environment, instead of traditionally reconstructing and optimizing the protein structure.
• On the prediction of higher-order mutations, the proposed strategy achieves remarkably higher performance by 20% improvement at the cost of requiring less than 1% of computational resources that famous transformer-based SOTA deep learning models for protein design require.

[Sep. 2021 - Apr. 2022] Medical Diagnosis: Integrating Convolutional Neural Networks and Graph Neural Networks [Paper] [Code]

Advised by Prof. Yuguang Wang

• Propose a fusion framework for enhancing the global image-level representation captured by CNNs with the geometry of cell-level spatial information learned by GNNs for downstream oncology tasks.
• Develop two fusion strategies optimizing an integration between collaborative features of global images and cell graphs, which can be generally applied to multiple networks.
• Evaluations on histology datasets have validated that fusion models outperform plain CNNs or GNNs and effectively reduce the cost of cancer diagnosis to design a stomach disease detection model.

[Sep. 2019 - Jun. 2021] Partial Differential Equations and Ordinary Differential Equations

Advised by Prof. Xiaoming Wang

• Investigate the modeling and numerical solutions of fluid partial differential equations.
• Build an infectious disease model based on multiple ordinary differential equations considering the effect of various factors.
• Predict the number of infected people and make predictions on COVID-19.

[Jun. 2018 - Oct. 2018] High-Resolution Photoacoustic Imaging

Advised by Prof. Lei Xi

• Generate high-resolution tissue images by measuring light-induced sound waves from its optically absorbing structures.

[Oct. 2021 - Mar. 2022] AI Software Intern

Taize Chiptech, Shanghai, China

• ResNet-18 on FPGA: Design essential forward propagation functions and stack them to run on FPGA.
• VLIW Compiler Development: Cooperate with fulfilling AI operator library in assembly language.
• Fixed Issue: ResNet-18 running results on FPGA were inconsistent with those generated by PyTorch.

Women's Soccer

• Member of the Women's Soccer Team of SJTU.
• Captain of the Women's Soccer Team of SUSTech Shuren College.
• WFAXSJTU Club Organizer: the first team in China to promote walking soccer, a sport similar to soccer but forbidden from running. We established the club to attract people with relative physical strength, such as senior citizens, to exercise more through soccer.