Brandon Wong
Education
Eindhoven University of Technology
Eindhoven, the Netherlands
Sep. 2024 - Jul. 2026 (Expected)
Research Honors Master of Science in Computer Science and EngineeringCurrent GPA 8.44/10.0
Northwest A&F University
Yangling, China
Sep. 2019 - Jul. 2024
Bachelor of Engineering in Software EngineeringMajor GPA 3.82/4.0
Graduation Thesis Research and Implementation of Skeleton Extraction Method for Incomplete Tree Trunk Point Clouds [Abstract]
Research
Generative Graph Diffusion over Maskable Abstract Syntax Trees
Sep. 2025 - May 2026 @ Eindhoven University of Technology
Supervised by Dr. Jelle Piepenbrock & Prof. Dr. Mark van den Brand
- Designed a neurosymbolic discrete diffusion framework to generate source code natively as Abstract Syntax Trees (ASTs).
- Engineered the Maskable AST (MAST) engine to strictly enforce context-free grammars, guaranteeing 100% syntactically valid code generation.
- Developed an R-GCN to guide the structural diffusion process by predicting graph topology refinements and terminal semantics.
- Formulated a structural evaluation methodology using Wasserstein distance and k-Motif distributions to measure macroscopic and microscopic topological fidelity.
- Outperformed token-based Transformer baselines in compilation rates and learning efficiency under strictly aligned parameter and training budgets.
Topology-aware Temporal Matching of River Channel Networks
Feb. 2025 - Aug. 2025 @ Eindhoven University of Technology
Supervised by Dr. Willem Sonke, Dr. Tim A.E. Ophelders & Dr. Kevin Verbeek
- One of two student researchers working on temporal matching of braided river evolution across multiple decades.
- Built upon Kleinhans et al.’s δ-network construction using descending quasi Morse–Smale complexes on riverbed terrains.
- Proposed a new hierarchical graph-based representation of braided river networks, enabling flow-consistent direction annotation of river segments.
- Integrated both geometric and topological context into segment matching, applying Dynamic Time Warping (DTW) to align river fragments across time.
- Refined DTW results with topology-aware correction, yielding more robust and interpretable temporal correspondences between river networks.
Context-Enriched Abstract Syntax Tree Construction for Code Representation
Apr. 2025 - Jul. 2025 @ Eindhoven University of Technology
Supervised by Dr. Jelle Piepenbrock
- Designed a multi-stage pipeline to transform raw C++ and Python source code into enriched Abstract Syntax Trees (ASTs) with unified structural semantics.
- Implemented type annotation extraction, variable reference resolution, and statement-order encoding to capture both syntactic and semantic program contexts.
- Developed language-specific enrichment strategies using
ClangASTfor C++ andastroid/astypesfor Python, harmonizing heterogeneous AST formats. - Introduced virtual edges for execution sequence and reference links to enable static validation of variable usage and control-flow consistency.
- Achieved high structural coverage and semantic completeness on large-scale datasets, producing ASTs suitable for structure-aware generative model training and cross-language analysis.
Automatic Stem Reconstruction from Incomplete 3D Point Cloud
Sep. 2022 - Jun. 2024 @ Northwest A&F University
Supervised by Dr. Shaojun Hu
- Designed and implemented a full pipeline for tree trunk modeling from incomplete LiDAR point clouds, including segmentation, skeleton extraction, optimization and reconstruction.
- Proposed an improved L1-medial skeleton method with RANSAC-based local cylinder fitting and global continuity correction, significantly enhancing robustness under occlusions.
- Developed a circle-constrained B-spline reconstruction model, enabling accurate and realistic stem surface recovery from incomplete cross-sections.
- Conducted quantitative and qualitative experiments on synthetic and real-world data, achieving up to 74.9% accuracy improvement in extreme missing-data scenarios.
- Contributed a practical solution for precision forestry, advancing reliable trunk modeling and measurement under challenging field conditions.
3D Animation of Simulated Butterfly Swarm Flight Based on Gesture Control
Oct. 2022 - May. 2024 @ Northwest A&F University
Supervised by Prof. Dr. Meili Wang
- Developed a graph-based control framework for swarm flight animation, preserving structural relations among individuals while following user-specified paths.
- Implemented path generation algorithms from both continuous curves and discrete gesture points, ensuring smooth swarm trajectories via Frenet–Serret frames and Catmull–Rom splines.
- Designed a tree-based hierarchical representation of swarm interior patterns, enabling pattern preservation and topological consistency during motion propagation.
- Proposed multiple motion propagation strategies (observation-based, latency-compensated, communication-based) and analyzed their trade-offs in accuracy, naturalness, and computational efficiency.
- Demonstrated that the framework balances biological realism and precise coordination, offering applications in computer animation, swarm robotics, and interactive VR environments.
Feature Point Extraction from 3D Point Clouds of Power Pylons
Sep. 2021 - Apr. 2023 @ Northwest A&F University
Supervised by Dr. Shaojun Hu
- Proposed a genetic algorithm–based axis correction method to normalize inclined pylon point clouds.
- Applied statistical segmentation to divide pylons into head, body, and base for segment-specific feature extraction.
- Combined geometric algorithms (Delaunay triangulation, Douglas–Peucker, RANSAC, Hough transform) with AI-inspired methods to improve robustness under noise and missing data.
- Achieved more complete and detailed 3D reconstruction of lattice structures compared to manual or purely model-driven approaches.
Publication
Graph-based control framework for motion propagation and pattern preservation in swarm flight simulations
May 2024
Authors: Feixiang Qi, Bojian Wang, Meili Wang
Computer Animation and Virtual Worlds, Volume 35, Issue 3