This project implements a Darwin-Gödel Machine (DGM) self-improvement framework to optimize class weights for imbalanced bridge damage classification. The system uses LLM-based coding agents to iteratively propose weight configurations, evaluated through LoRA fine-tuning on a Japanese BERT-large model.
hyperparameter-optimization lora imbalanced-learning structural-health-monitoring experiment-automation bridge-maintenance loss-function-optimization class-weighting coding-agent darwin-godel-machine self-improvement-ai bert-large-japanese-v2 automated-code-optimization damage-cause-encoder
-
Updated
Aug 12, 2026 - Python