Competition on LLM Designed EA, GECCO

Time: Mar 2026

• Collaborated in a team to design adaptive multi-operator evolutionary algorithms using the LLM4AD platform and Evolution of Heuristics (EoH).
• Designed a module based on the Multi-Armed Bandit (MAB) algorithm for adaptive operator selection during optimization.
• Achieved optimal results on 17/24 GNBG benchmark problems.

Research on Combinatorial Optimization Methods Based on Deep Learning and Development of the EasyCO Platform

Time: Sep 2025 - Present

• Explored graph neural network based data representations and feature learning mechanisms integrated with reinforcement learning frameworks (for example REINFORCE).
• Contributed to the open-source EasyCO platform for building general-purpose efficient solvers for industrial applications.

Autonomous Driving Video Annotation System Based on VideoLLaMA2

Time: Apr 2025 - Jun 2025

• Designed English prompts focused on key driving elements to generate descriptions centered on ego-vehicle behavior and interactions with preceding vehicles.
• Fine-tuned the model to optimize accuracy.