Experience

Core Experimental Contributor

Concept-Based Dictionary Learning for Inference-Time Safety in VLA Models

  • Core Experimental Contributor (CVPR under review).
  • Architected and implemented a modular framework for concept-dictionary–based inference-time safety defenses in Vision–Language–Action (VLA) models.
  • Implemented reusable components for concept extraction, intervention hooks at inference time, and end-to-end evaluation scripts.
  • Curated and constructed evaluation datasets to stress-test safety and generalization, including adversarial and OOD scenarios.
  • Led systematic ablation studies to isolate the impact of dictionary size, concept selection strategies, and intervention strength.
  • Executed large-scale experiments across simulated and real-world-like settings, ensuring reproducibility via configuration management.

Team Member

ICPC-Style Programming Contests

  • Regularly participate in ICPC-style contests, solving algorithmic problems in C++ under tight time constraints.
  • Practice spans graph algorithms, dynamic programming, greedy methods, data structures, and basic computational geometry.

Education

BSc Computer Science (Joint Programme)

Beijing Jiaotong University & Lancaster University

Year 2. GPA: 3.71 / 4.0
Skills
Technical Skills
C/C++
Python (PyTorch, NumPy)
Java
Git/Linux/Docker
LaTeX
Deep Learning
Model Training & Evaluation
Dataset Preprocessing
Ablation Studies
Awards
Bronze Medal
Weihai Municipal University Students Programming Contest ∙ Present